Circuits and Methods of Generating and Controlling Signals on an
Integrated Circuit
CROSS REFERENCE TO RELATED APPLICATIONS [0001] This invention claims the benefit of priority from U.S. Patent Application No. 11/038,560, filed January 19, 2005, entitled "Fully Integrated Frequency Generator," the disclosure of which, is hereby incorporated herein by reference. This invention claims the benefit of priority from U.S. Provisional Application No. 60/645,837, filed January 21, 2005, entitled "Controlled Delay Line Circuit with Integrated Transmission Line Reference," the disclosure of which, is hereby incorporated herein by reference. This invention claims the benefit of priority from U.S. Provisional Application No. 60/663,655, filed March 31, 2005, entitled "Fully Integrated Frequency Generator," the disclosure of which, is hereby incorporated herein, by reference. This invention claims the benefit of priority from U.S. Patent Application ISTo. 11/200,472, filed August 9, 2005, entitled "Circuits and Methods for Reducing Static Phase Offset using Commutating Phase Detectors," the disclosure of which is hereby incorporated herein by reference. This invention claims the benefit of priority from U.S. Patent Application No. UNASSIGNED (Attorney Docket No. 000012-00020 IUS), filed January 11, 2006, entitled "Controlled Delay Line Circuit with Integrated Transmission Line Reference," the disclosure of which is hereby incorporated herein by reference. BACKGROUND
[0002] The present invention relates to generating and controlling signals on an integrated circuit. One embodiment of the present invention relates to frequency generators, and in particular, to frequency generator circuits and methods that can be fully integrated on a single integrated circuit. [0003] In almost every computer system or digital system, there is a requirement for at least one signal having a specified frequency. One example of such a signal is a reference frequency signal (i.e ., a timing reference). For example, a reference frequency signal or timing reference may be a system clock. Signals with specified frequencies, such as a system clock, are created using frequency generators. One example of a frequency generator is a crystal oscillator. A crystal oscillator is an electrical circuit that uses a quartz crystal as a reference device to generate a frequency. Quartz crystals are piezoelectric devices that are made from crystalline silicon dioxide. When quartz crystals are driven by an electrical signal, they will exhibit a mechanical resonance (vibration) at certain frequencies of the
driving signal. By using the appropriate electrical circuit, an electrical signal can be generated that is equal in frequency to the quartz crystal's mechanical resonant frequency. Such circuits are advantageous because quartz crystal may be used to generate very precise reference frequency signals. [0004] Even though crystal oscillators have an advantage of being very accurate, they do have some well-known disadvantages. For example, since the creation of the reference frequency involves physically vibrating a silicon dioxide crystal at the reference frequency, over-driving the crystal with too large of an electrical signal can damage it. This damage to the quartz crystal can result in a shift of the resonant frequency, or in extreme cases the crystal can fracture. If a fracture were to occur, the crystal would become non-functional and the crystal oscillator would stop operating at the reference frequency. Sudden large changes in temperature can also damage the quartz crystal. Again, this damage could result in a shift of the crystal's resonant frequency or the fracture of the crystal. Another disadvantage associated with these crystal oscillators are their susceptibility to mechanical vibration or shock. A mechanical shock to the crystal can cause a sudden momentary shift in the oscillator frequency. This occurs because the stiock can disturb the mechanical vibration of the quartz crystal. In the same manner, a constant mechanical vibration of the circuit board to which the crystal is attached can interfere with the mechanical vibration of the quartz crystal. A mechanical vibration of the circuit board would, cause periodic variations of the output frequency of the crystal oscillator.
[0005] A particularly significant drawback to crystal oscillator circuits is that a quartz crystal cannot be integrated into a monolithic integrated circuit together with the electrical drive circuitry. The silicon dioxide crystal is always placed external to the integrated circuit, which contains the electrical devices that drive tibe crystal. Because the crystal is external to the integrated circuit, the crystal oscillator is much more susceptible to electrical disturbances from external sources. Signals adjacent to the quartz crystal can couple electrical disturbances into the leads of the crystal. These disturbances may result in variations of the output frequency of the oscillator circuit. Having the quartz crystal external to the integrated circuit may also result in the crystal oscillator being more susceptible to humidity and dirt. Accumulated moisture or dirt across the leads of the crystal would create a conduction path between the leads. If the resistance of this parasitic conduction path becomes too low, the crystal oscillator circuit would stop oscillating. Still another drawback to the crystal oscillator is related to economics. Since the quartz crystal is external and separate from the
integrated circuit that drives it, additional costs are incurred. One additional cost is due to the extra assembly costs required to attach the external crystal to the printed circuit board. Another additional cost is due to the extra printed circuit "board space that is used.
[0006] Despite these many disadvantages, crystal oscillators are very popular because they are capable of generating extremely accurate reference frequency signals. The precision of the output frequencies from these crystal oscillator circuits is on the order of 0.01%. However, in many applications such a high level of precision is not required. For example, in many digital systems a precision on the order of 1.0% for the system clock is sufficient.
[0007] Thus, there is a need for improved frequency signal generators over existing crystal oscillator techniques. In particular, there is a need for a frequency generator that is fully integrated on a single integrated circuit. The present invention solves these and other problems by providing a frequency generator that can be fully integrated on a single integrated circuit. Features and advantages of the present invention include providing a frequency generator with improved reliability and lower cost. [0008] The present invention also relates to controlled delay lines and delay locked loops, and in particular, to a controlled delay line circuit with an integrated transmission line reference.
[0009] One of the most important issues in the design of the digital logic section of an integrated circuit is the system clock to data timing relationship. In order to optimize this important timing relationship, a controlled delay line is often used. The controlled delay line is used to delay the data and/or the system clock in such a way as to improve the timing relationship. For example, one common type of controlled delay line that is often used is a voltage controlled delay line (VCDL). Voltage controlled delay lines are advantageous because the delay time of a VCDL is relatively accurate, predictable, and stable. The accuracy and stability of the voltage controlled delay line depends on how the control voltage for the VCDL is generated. The control voltage is typically^ created using a reference VCDL within the feedback loop of a Delay Locked Loop (DLL). The DLL forces the reference VCDL to have the same time delay as the timing reference for the DLL. The timing reference for the DLL is usually the period of the system clock or some multiple of this period. Since the system clock is normally oscillating at a very precise and stable frequency, this results in the delay time of the VCDL also being relatively precise and stable.
[0010] In some applications, there is a requirement that the digital logic of an integrated circuit has to operate at more than one system clock frequency. If the system clock shifts to a new frequency, the delay time of the VCDL could also change if it is dependent on the period of the system clock. This change in delay time for the VCDL may adversely affect the system clock to data timing relationships . For this design situation, a voltage controlled delay line that is independent of the system clock would be advantageous.
[0011] One possible method to create a VCDL that is not dependent on the system clock is to use a resistor and capacitor network as a timing reference device for the Delay Locked Loop. Even though a resistor and capacitor network can be integrated into an integrated circuit, the repeatability and stability of an integrated resistor and capacitor network's RC time constant is poor. An integrated circuit RC time constant can vary quite a lot due to variations in the manufacturing process. IFor example, such manufacturing process variations may cause the value of an integrated resistor to vary as much as +/-20% and the value of an integrated capacitor can vary as much as -+/-10%. These large variations in the values of the integrated resistors and capacitors mean ttiat the precision of an integrated circuit RC time constant is low. Moreover, the value of an integrated resistor can change significantly over a temperature range. The value of an integrated resistor will typically change by 7% over a temperature range of 70 degrees Celsius. Thus, the RC time constant of an integrated circuit resistor and capacitor network is not very stable across a temperature range. Because of the wide changes in value due to manufacturing and temperature variations, an integrated circuit resistor and capacitor network would not make a good timing reference for an integrated DLL.
[0012] Thus, there is a need for an improved controlled delay line circuit that is repeatable, stable across manufacturing processes, and. independent of the system clock. [0013] The present invention also relates to reducing static phase offset in timing loops, and in particular, to circuits and methods of reducing static phase offset using commutating phase detectors.
[0014] Timing loops are feedback loop circuits that are used in a wide variety of electronic applications. Timing loops are typically used to generate signals that have particular frequencies, periods, or delays. Such loops include phase locked loops and delay locked loops, for example. Fig. IA illustrates phase locked loop architecture, which is one type of timing loop. Phase locked loop ("PLL") IOOA includes a phase detector 110, loop filter 120,
voltage controlled oscillator ("VCO") 130, and may optionally include feedback 140 such as a divider, for example. Phase detector 110 includes two input terminals and an output terminal. The First input terminal receives a reference frequency ("REF"), which may be a digital or analog signal. The second input terminal of phase detector 110 is a feedback signal ("FB") coupled to the output of VCO 130. Phase detector 110 translates phase differences between the input signals into an output signal. The output of phase detector 110 may be a current into a capacitance in loop filter 120, for example. The ouiput of loop filter 120 is a voltage that controls the frequency of oscillation of VCO 130. Tfcie output frequency generated by VCO 130 may be divided by feedback 140 or coupled directly to the feedback input of phase detector 110.
[0015] As mentioned above, timing loops are used in a variety of applications. For example, in the design of computer systems, maintaining adequate timing margin between the computer system's clock signal and data is very important. As the frequency of the system clock increases, maintaining adequate timing margin becomes even more difficult. Thus, the design of clock distribution networks for the computer becomes nxore and more difficult as frequencies increase. One application of a timing loop is a zero delay buffer ("ZDB"), which is shown in Fig. IB. In ZDB 10OB, the output of the VCO is typically fed back directly to the input of the phase detector. In this example, the phase detector is implemented using a phase frequency detector and a charge pump. A ZDB may be used to improve the performance of the clock distribution network in a computer system. A ZDB may be used to regenerate a clock signal to improve drive capability or to regenerate multiple copies of the system clock. An ideal ZDB has an output that is an identical version of the input signal in phase and frequency (e.g., the divider modulus of the feedback shown in Fig. IA is equal to one). Since the output signal of an ideal ZDB is exactly in phase with the input signal, there would appear to be no delay in the buffer, hence the name "zero delay buffer." However, in a real application some delay is introduced by non-idealities in the PLL. For example, "static phase offset" is a specification that quantifies the phase difference between the input signal and the output signal. Static phase offset is the average phase offset between the input reference signal received by the PLL and the output of the VCO. In other words, static phase offset is the difference in time between the input and the output signals. Ideally, the static phase offset should be zero seconds for a ZDB. However, in real world applications there is always some phase error between the input and output.
[0016] Fig. 1C illustrates another timing loop. Timing loop IOOB is delay locked loop ("DLL") architecture, which is another type of timing loop. An example delay locked loop includes a phase detector 110, filter 120, and voltage controlled delay 150. Am input signal to be delayed is coupled to one input of the voltage controlled delay 150, and a second input of the voltage controlled delay is coupled to the output of phase detector 110. Tbie input signal will be delayed by an amount of time set by a voltage at the output of phase detector i 10. The input signal to be delayed is also coupled to one input of phase detector 1 10, and the output of the voltage controlled delay 150 is coupled to the other input of the phase detector 110. In some applications feedback (such as a divider) may be included between the output of the voltage controlled delay and the input of phase detector 110. Thus, in trαe absences of static phase offset error, the output of the controlled delay 150 will be precisely in phase with the input signal when the loop is closed.
[0017] Typically, the majority of static phase offset error is generated in the phase detector. For example, referring to Fig. IA, if the delay between REF input of phase detector 110 to the output of the phase detector is the same as the delay from FB input of phase detector 110 to the output then there should be no static phase offset. Similarly, if the circuitry in the phase detector is matched, then there should be no static phase offset. For example, if the phase detector includes a phase frequency detector and charge pump, then if the pull up and pull down currents in the charge pump are matched, then there should be no sta.tic phase offset. However, typically the delays are different and the circuitry is mismatchied due to manufacturing variations and imperfections.
[0018] Static phase offset is problematic in timing loop designs. In particular, static phase offset in timing applications can severely impact the timing margins, such as computer system clock margins described above. Thus, it would be desirable to reduce static phase offset so that the timing margin in computer system clocks can be improved. Siαch improved timing margins would allow for greater and more robust system performance.
[0019] Thus, there is a need for reducing static phase offset in timing loops. TTie present invention solves these and other problems by providing circuits and methods for* reducing static phase offset in timing loops using commutating phase detectors. SUMMARY
[0020] The present invention relates to generating and controlling signals on an integrated circuit. One embodiment of the present invention relates to frequency generators, and in
particular, to frequency generator circuits and methods that can be fully integrated on a single integrated circuit. In one embodiment, the present invention includes an integrated circuit feedback loop comprising an integrated transmission line, wherein the integrated transmission line is used as a timing reference component for the feedback loop and wherein the feedback loop and the transmission line are integrated on a single integrated circuit. The present invention has a variety of uses. In one embodiment, the present invention may be used as a frequency generator comprising a feedback loop with a transmission line integrated on a single integrated circuit.
[0O21] In one embodiment, a frequency generator may include a phase detector and a voltage controlled oscillator coupled in series, and a transmission line having an input coupled to an output of the voltage controlled oscillator, the transmission line providing a time delay between the transmission line input and a transmission line output, wherein the phase detector includes an input coupled to an output of the voltage controlled oscillator and another input coupled to the transmission line input. Foe example, in one embodiment, the one input of the phase detector may be coupled to the input of the transmission line. The phase detector, voltage controlled oscillator and transmission line are advantageously integrated on a single integrated circuit.
[0022] In another embodiment, the present invention includes a controlled delay line circuit comprising a feedback loop including an integrated transmission line, wherein the integrated transmission line is used as a timing reference for the feedback loop and wherein the feedback loop and the transmission line are integrated on a single integrated circuit. The feedback loop may be a Delay Locked Loop, for example.
[0023] In one embodiment, the present invention includes an integrated circuit comprising a first controlled delay line coupled in a feedback loop and an integrated transmission line, and the integrated transmission line is used as a timing reference for the feedback loop and wherein the feedback loop and the transmission line are integrated on a single integrated circuit.
[0024] In one embodiment, the feedback loop includes a phase detector having a first input coupled to an output of the integrated transmission line and a second input coupled to an output of the first controlled delay line.
[0025] In one embodiment, the first controlled delay line and the integrated transmission line have inputs that are coupled to a circuit that generates a first periodic signal.
[0026] In one embodiment, the circuit further comprises a phase-frequency detector having a first input coupled to an output of the integrated transmission line and a second input coupled to an output of the first controlled delay line, a charge pump coupled to an output of the phase- frequency detector, and a capacitor coupled to the output of the charge pump and to a control input of the first controlled delay line.
[0027] In one embodiment, the circuit further comprise a second controlled delay line having a control input coupled to the control input of the first controlled delay line.
[0028] In one embodiment, the second controlled delay line receives an input signal, and generates a delayed version of said input signal. [0029] In. one embodiment, the present invention includes an integrated circuit comprising an integrated transmission line having an input coupled to receive a periodic signal, the transmission line delaying the periodic signal by a first delay to produce a first delayed signal, a first controlled delay line having a first input coupled to the periodic signal and a first control input, the first controlled delay line delaying the periodic signal by a first variable delay to produce a second delayed signal, a phase detector having a first input coupled to an output of the integrated transmission line and a second input coupled to an output of the first controlled delay line, and a filter coupled between an output of the phase detector and the control input of the first controlled delay line.
[0030] In one embodiment, the phase detector, the first controlled delay line, and the transmission line are integrated on a single integrated circuit.
[0031] In one embodiment, the integrated transmission line comprises one or more conductive interconnect layers.
[0032] In one embodiment, wherein the filter is a capacitor.
[0033] In one embodiment, the circuit further comprises a second controlled delay line having a control input coupled to the filter.
[0034] In one embodiment, the phase detector is a phase-frequency detector coupled to a charge pump.
[0035] In one embodiment, the periodic signal is generated by an oscillator circuit, and the phase-frequency detector includes an enable input coupled to the oscillator circuit.
[0036] In one embodiment, the present invention includes a method of controlling the delay of a signal comprising generating a first signal, delaying the first signal by a first delay in an integrated transmission line to produce a first delayed signal, and delaying the first signal by a first variable delay in a first controlled delay line to produce a second delayed signal, wherein the first controlled delay line is configured in a feedback loop, wherein the integrated transmission line is used as a timing reference for the feedback loop to adjust the first variable delay to be equal to the first delay, and wherein the feedback loop and the transmission line are integrated on a single integrated circuit.
[0037] In one embodiment, the method further comprises delaying a second signal by a second variable delay in a second controlled delay line.
[0038] In one embodiment, the second variable delay is equal to the first variable delay.
[0039] In one embodiment, the method further comprises detecting a phase difference between the first delayed signal and the second delayed signal.
[0040] In one embodiment, the method further comprises increasing the first variable delay if the first delay is greater than the first variable delay, and decreasing the first variable delay if the first delay is smaller than the first variable delay.
[0041] In one embodiment, a time difference between the first delayed signal and the second delayed signal is translated into a voltage on a capacitor and used to control the first variable delay. [0042] In one embodiment, the method further comprises generating a first pulse in response to the first delayed signal and generating a second pulse in response to the second delayed signal, discharging a current from a capacitor in response to receiving the first pulse, and charging a current into a capacitor in response to receiving the second pulse, and in accordance therewith, changing a first voltage on the capacitor, and adjusting the first variable delay using the first voltage on the capacitor.
[0043] In one embodiment, the present invention includes an integrated circuit comprising transmission line means for delaying a periodic reference signal means by a fixed amount of time, controlled delay line means for delaying the periodic reference signal means by a first variable amount of time, and feedback loop means for adjusting the first variable amount of time to equal the fixed amount of time, wherein the transmission line means, the controlled delay line means, and the feedback loop means are integrated on a single integrated circuit.
[0044] Embodiments of the present invention may also reduce static phase offset in timing loops. In one embodiment, the present invention includes a timing loop comprising first and second phase detectors, wherein during a first time period, the first phase detector is coupled in a closed timing loop and the second phase detector is decoupled from the closed timing loop and calibrated, and during a second time period, the second phase detector is coupled in a closed timing loop and the first phase detector is decoupled from the closed timing loop and calibrated.
[0045] In one embodiment, when the first or second phase detector is calibrated, first and second inputs of the phase detector being calibrated are coupled to a common input signal. [0046] In one embodiment, when the first or second phase detector is calibrated, at least one output of the phase detector being calibrated is coupled to a mismatch detector for detecting mismatches in the phase detector.
[0047] In one embodiment, when the first or second phase detector is calibrated, at least one output of the phase detector being calibrated is coupled to a mismatch detector for detecting a voltage drift at the output of the phase detector that is indicative of the mismatches in the phase detector.
[0048] In one embodiment, the first phase detector includes a first input, a second input, an adjust input, and an output, the second phase detector includes a first input, a second input, an adjust input, and an output, a voltage controlled oscillator having an input coupled to the output of the first phase detector and the output of the second phase detector, and an output coupled to the second input of the first phase detector and the second input of the second phase detector, and a mismatch detector having one or more inputs coupled to the output of the first phase detector and the output of the second phase detector, and one or more outputs coupled to the adjust input of the first phase detector and the adjust input of the second phase detector, wherein during the first time period, the output of the first phase detector is coupled to the voltage controlled oscillator, the first and second inputs of the second phase detector are coupled to a common input signal, and the output of the second phase detector is coupled to the mismatch detector, and in accordance therewith, the mismatch detector generates a correction signal for adjusting the second phase detector, and during a second time period, the output of the second phase detector is coupled to the voltage controlled oscillator, the first and second inputs of the first phase detector are coupled to a common input signal, and the output of the first phase detector is coupled to the mismatch detector, and in accordance
therewith, the mismatch detector generates a correction signal for adjusting the first phase detector.
[0049] In one embodiment the first phase detector includes a first input, a second input, an adjust input, and an output, the second phase detector includes a first input, a second input, an adjust input, and an output, a voltage controlled delay having a first input coupled to the output of the first phase detector and the output of the second phase detector, a second input coupled to an input signal to be delayed, and an output coupled to the second input of the first phase detector and the second input of the second phase detector, and a mismatch detector having one or more inputs coupled to the output of the first phase detector and the output of the second phase detector, and one or more outputs coupled to the adjust input of the first phase detector and the adjust input of the second phase detector, wherein during the first time period, the output of the first phase detector is coupled to the voltage controlled delay, the first and second inputs of the second phase detector are coupled to a common input signal, and the output of the second phase detector is coupled to the mismatch detector, and in accordance therewith, the mismatch detector generates a correction signal for adjusting the second phase detector, and during a second time period, the output of the second phase detector is coupled to the voltage controlled delay, the first and second inputs of the first phase detector are coupled to a common input signal, and the output of the first phase detector is coupled to the mismatch detector, and in accordance therewith, the mismatch detector generates a correction signal for adjusting the first phase detector.
[0050] In one embodiment, the first and second phase detectors each comprise a phase frequency detector and a charge pump.
[0051] In one embodiment, the first and second phase detectors each further include an adjust input, wherein the adjust input of the first phase detector adjusts at least a first current in a first charge pump, and the adjust input of the second phase detector adjusts at least a second current in the second charge pump.
[0052] In one embodiment, the timing loop further comprises a first calibration capacitor coupled to the first phase detector during the second time period, and a second calibration capacitor coupled to the second phase detector during the first time period. [0053] In one embodiment, the timing loop further comprises a window comparator coupled to an output of the second phase detector during the first time period and coupled to an output of the first phase detector during the second time period.
[0054] In one embodiment, the timing loop further comprises a current pulse generator coupled to an output of the window comparator.
[0055] In one embodiment, the timing loop further comprises a phase detector configuration controller, [0056] In one embodiment, the phase detector configuration controller comprises a counter.
[0057] In one embodiment, the phase detector configuration controller further comprises a phase frequency detector.
[0058] In one embodiment, the timing loop is a phase locked loop or a delay locked loop. [0059] In one embodiment, the timing loop is configured as a zero delay buffer. [0060] In another embodiment, the present invention includes a timing loop comprising first and second phase detectors alternately configured in a correction loop and a closed timing loop, wherein during a first time period, the first phase detector is configured in the closed timing loop and the second phase detector is configured in the correction loop, and during a second time period, the second phase detector is configured in the closed timing loop and the first phase detector is configured in the correction loop.
[0061] In one embodiment, first and second inputs of the second phase detector are coupled to a common input signal during the first time period, and first and second inputs of the first phase detector are coupled to a common input signal during the second time period.
[0062] In one embodiment, the first and second phase detectors each comprise a phase frequency detector and a charge pump.
[0063] In one embodiment, the correction loop detects mismatch in the first and second phase detectors and generates a correction signal, wherein the correction signal is coupled to an adjust input of the first phase detector to adjust a first current in a first charge pump during the second time period, and the correction signal is coupled to an adjust input of the second phase detector to adjust a second current in a second charge pximp during the first time period.
[0064] In one embodiment, the correction loop comprises a first calibration capacitor coupled to the first phase detector during the second time period and a second calibration capacitor coupled to the second phase detector during the first time period.
[0065] In one embodiment, the correction loop comprises a window comparator.
[0O66] In one embodiment, the correction loop further comprises a current pulse generator.
[0O67] In one embodiment, the timing loop further comprises a counter for controlling the duration of the first and second time periods. [0O68] In one embodiment, the timing loop further comprises a phase frequency detector coupled to an input of the counter.
[0069] In one embodiment, the timing loop is a phase locked loop or a delay locked loop. [0070] In one embodiment, the timing loop is configured as a zero delay buffer.
[0071] In another embodiment, the present invention includes a method comprising, during a first time period, configuring a first phase detector to control a timing loop and calibrating a second phase detector, and during a second time period, configuring the second phase detector to control the timing loop and calibrating the first phase detector.
[0072] In one embodiment, first and second inputs of the second phase detector are coupled to a common input signal during the first time period, and first and second inputs of the first phase detector are coupled to a common input signal during the second tune period.
[0073] In one embodiment, the timing loop is a phase locked loop or a delay locked loop. [0074] In one embodiment, the timing loop is configured as a zero delay buffer.
[0075] In one embodiment, during the first time period calibrating comprises detecting mismatches in the second phase detector, and during the second time period calibrating comprises detecting mismatches in the first phase detector.
[0076] In one embodiment, during the first time period calibrating comprises detecting a voltage drift at the output of the second phase detector, and during the second time period calibrating comprises detecting a voltage drift at the output of the first phase detector.
[0077] In one embodiment, during the first time period calibrating further comprises comparing the voltage drift at the output of the second phase detector to at least one reference voltage, and during the second time period calibrating further comprises comparing the voltage drift at the output of the first phase detector to at least one reference voltage.
[0078] In one embodiment, the first phase detector comprises a first phase frequency detector and a first charge pump, and the second phase detector comprises a second phase frequency detector and a second charge pump.
[0079] In one embodiment, during the first time period calibrating fuxther comprises adjusting a voltage on a gate of a first transistor coupled to one or both of a current source or current sink in the second charge pump, and during the second time period calibrating further comprises adjusting a voltage on a gate of a second transistor coupled to one or both of a current source or current sink in the first charge pump.
[0080] In yet another embodiment, the present invention includes a method comprising, during a first time period, configuring a first phase detector to control a timing loop, detecting a first mismatch in the second phase detector, and reducing the first mismatch generated by the second phase detector, during a second time period, configuring the second phase detector in a timing loop, detecting a second mismatch in the first phase detector, and reducing the second mismatch generated by the first phase detector. [0081] In one embodiment, first and second inputs of the second phase detector are coupled to a common input signal during the first time period, and first and second inputs of the first phase detector are coupled to a common input signal during the second time period.
[0082] In one embodiment, reducing includes adjusting the amount of current generated at an output of the first or second phase detectors to reduce the first or second mismatch. [0083] In one embodiment, the first phase detector comprises a first phase frequency detector and a first charge pump, and the second phase detector comprises a second phase frequency detector and a second charge pump.
[0084] In one embodiment, during the first time period, reducing comprises adjusting a voltage on a gate of a first transistor coupled to one or both of a current source or current sink in the second charge pump, and during the second time period reducing comprises adjusting a voltage on a gate of a second transistor coupled to one or both of a current source or current sink in the first charge pump.
[0085] In one embodiment, during the first time period, detecting comprises comparing a voltage drift at an output of the second phase detector to at least one reference voltage, and during the second time period detecting comprises comparing a voltage drift at an output of the first phase detector to at least one reference voltage.
[O086] The following detailed description and accompanying drawings provide a better understanding of the nature and advantages of the present invention.
BRIEF DESCRIPTION OF THE DRAWINGS [0087] Fig. IA illustrates phase locked loop architecture. [0088] Fig. IB illustrates a zero delay buffer. [0089] Fig. 1C illustrates a delay locked loop.
[0090] Fig. 2 illustrates an integrated circuit according to one embodiment of the present invention.
[0091] Fig. 3 illustrates a frequency generator according to another embodiment of the present invention.
[0092] Fig. 4 illustrates a frequency generator according to another embodiment of the present invention.
[0O93] Fig. 5 is an example of a phase detector according to one embodiment of the present invention. [0O94] Fig. 6 is an example waveform to illustrate pulse removal according to one embodiment of the present invention.
[0O95] Fig. 7 is a timing diagram to illustrate the operation of a frequency generator according to one embodiment of the present invention.
[0O96] Fig. 8 is another timing diagram to illustrate the operation of a frequency generator according to one embodiment of the present invention.
[0097] Fig. 9 illustrates a frequency generator according to anottier embodiment of the present invention.
[0098] Fig. 10 is a timing diagram to illustrate the operation of tb.e frequency generator of Fig. 8 according to one embodiment of the present invention. [0099] Fig. 11 is another timing diagram to illustrate the operation of the frequency generator of Fig. 8 according to one embodiment of the present invention.
[0100] Fig.l2A illustrates an integrated transmission line used as a timing reference in a feedback loop according to one embodiment of the present invention.
[0101] Fig.12B illustrates a controlled delay line with an integrated transmission line reference according to one embodiment of the present invention.
[0102] Fig. 13 is an example of a phase detector according to one embodiment of the present invention. [0103] Fig. 14 is a timing diagram to illustrate the operation of a controlled delay line according to one embodiment of the present invention.
[0104] Fig. 15 is another timing diagram to illustrate the operation of a controlled delay line according to one embodiment of the present invention.
[0105] Fig. 16A illustrates a timing loop according to one embodiment of the present invention.
[0106] Fig. 16B illustrates a timing loop according to another embodiment of the present invention.
[0107] Fig. 17 is an example application of a phase locked loop with reduced static phase offset according to one embodiment of the present invention. [0108] Fig. 18A-B is an example of an auto-zero phase detector according to one embodiment of the present invention.
[0109] Fig. 19 is an example of an adjustable charge pump according to one embodiment of the present invention.
[0110] Fig. 20 is an example of a phase-frequency detector according to> one embodiment of the present invention.
[0111] Fig. 21 is an example of a bias generator according to one embodiment of the present invention.
[0112] Fig. 22 shows example waveforms for a circuit according to one embodiment of the present invention. DETAILED DESCRIPTION
[0113] Described herein are techniques for generating and controlling signals generating reference frequencies on an integrated circuit. Circuits and methods are disclosed for generating reference frequencies and controlled delays on an integrated circuit. The present disclosure also includes circuits and methods that may be used to implement phase detection,
which may be used to improve the accuracy of the frequency generation and delay controls disclosed herein. In the following description, for purposes of explanation, numerous examples and specific details are set forth in order to provide a thorough understanding; of the present invention. It will be evident, however, to one skilled in the art that the present invention as defined by the claims may include some or all of the features in these examples alone or in combination with other features described below, and may further include ofcvious modifications and equivalents of the features and concepts described herein.
[0114] Fig. 2 illustrates an integrated circuit 200 according to one embodiment of the present invention. Integrated circuit 200 includes a feedback loop comprising a feed forward path 210 and an integrated transmission line 220. The circuitry for the feed forward path 210, integrated transmission line 220 and any other application specific circuitry are integrated on a single integrated circuit. Embodiments of the present invention use the integrated transmission line 220 as a timing reference for the feedback loop. Thus, integrated circuit 200 may be used in a variety of applications that require a timing reference, such as a frequency generator or controlled delay, for example.
[0115] Circuits used in timing applications, such as frequency generator circuits or delay lines, require some form, of reference device to set the time base of the signals (i.e., a tuning reference or frequency reference). A good reference device should be accurate, repeatab>le, and predictable. Moreover, such devices should be stable with temperature changes and power supply voltage changes. Developing a good timing or frequency reference device that can be integrated into an integrated circuit is extremely difficult. A quartz crystal, for example, cannot be built into an integrated circuit, but rather, must reside external to the integrated circuit that drives it. Typically, in an integrated circuit the RC time constant of a resistor and capacitor network has been used as a timing reference. Resistors and capacitors can be built into an integrated circuit, but an integrated circuit RC time constant can exhitit unacceptable variations due to variations in the manufacturing process. For example, the value of an integrated resistor can vary as much as +/-20%, and the value of an integrated! capacitor can vary as much as +/-10% because of variations in the manufacturing process . Thus, the time constant of an integrated resistor and capacitor network is not very repeata.t>le. Moreover, the value of an integrated resistor can change significantly over a temperature range. The value of an integrated resistor will typically change by 7% over a temperature range of 70 degrees Celsius. Therefore, the RC time constant of an integrated circuit resistor and capacitor network is not stable with temperature. Because of the wide changes in value
due to manufacturing and temperature variations, an integrated circuit resistor and capacitor network does not make a good timing reference device.
[0116] Embodiments of the present invention include building an integrated transmission line (ITL) into an integrated circuit as the timing reference device. The flight time delay of the transmission line may be used as a timing reference for a feedback loop in a frequency generator, phase locked loop, or delay locked loop, for example, that is integrated on trie same chip. Time delay in a transmission line is caused by the propagation of electromagnetic waves down the transmission line. Such time delays may further include some delay caused by parasitic resistances of the delay line. Embodiments of the present invention may preferably minimize the delay effects of parasitic resistances because such resistances may cause the time delay of an integrated transmission line to vary across process and temperature. More generally, the flight time delay of a transmission line depends on a variety of parameters, including geometric shape, length, construction, and the relative dielectric constant of the transmission, line's insulator. One common transmission line is stripline. The flight time delay for a stripline transmission line can be calculated from the following equation: ττL = (LTL (εr)1/2 ) / c0 where,
TTL = flight time delay of the transmission line LTL = length of the transmission line
&r = relative dielectric constant of the transmission line insulator. C0 = the speed of light in a vacuum.
These equations apply to transmission lines with fields contained in a single dielectric. Examples of such transmission lines include stripline and coax. However, these equations are illustrative only. The present invention may use any integrated transmission line that produces a stable time delay, such as microstrip, for example, and others.
[0117] Stability and repeatability advantages of using an integrated transmission line Ln a frequency generator can be seen by examining the terms of the flight time delay equation. The length of the integrated transmission line, LTL, may be set by design choice. Once the
value of the term LTL is designed, an integrated transmission line will have negligible changes in length due to variations in the manufacturing of the integrated circuit.
[0118] With regard to the other parameters effecting flight time delay, the term c0 is the speed of light in a vacuum, which is a physical constant that is invariable. The value of the term B1 depends on the dielectric material used as the insulator for the integrated transmission line. In the current manufacturing environment, the most common insulator for a transmission line on an integrated circuit would be amorphous silicon dioxide. The relative dielectric constant, εr, for silicon dioxide does not change with manufacturing variations and exhibits only a very small variation due to temperature changes. The temperature coefficient for silicon dioxide's relative dielectric constant is approximately 20ppm/°C. The variation of εr due to temperature change translates into a temperature coefficient for TTL of approximately 10ppm/°C. For a 7O0C change in temperature, the total variation of TTL should be less than 0.1% due to the variation in εr.
[0119] Another advantage of an integrated transmission line is that the time delay of an integrated transmission line is not dependent on the magnitude of the signal that is propagating along the transmission line. Thus, the time delay, τTL, is independent of the power supply voltage of the integrated circuit. Therefore, because the flight time delay of a transmission line is independent of the power supply voltage, exhibits very good stability versus temperature change, and changes very little due to manufacturing variations, the flight time delay of an integrated transmission line would make a good timing reference for an integrated, feedback loop, such as a phase locked loop or DLL. For example, using the integrated, transmission line as a timing reference, it is possible to create a frequency generator that has a total frequency variation of less than +/- 1.0%. This total variation would include all of the variations due to manufacturing, temperature, and power supply voltage. Therefore, embodiments of the present invention advantageously include both a transmission line (to act as a timing reference) and other frequency generator circuitry on the same integrated, circuit. Additionally, some embodiments of the present invention advantageously include a voltage controlled delay line with an integrated transmission line as the timing reference. [0120] In some embodiments, the present invention, may use one or more conductive interconnect layers (e.g., metalization) for constructing an integrated circuit transmission line. For example, connecting multiple metalization layers in parallel can reduce the resistance of
the integrated transmission line, resulting in an improved transmission line. One common metalization material is aluminum. Additionally, other processes may use one or more copper metalizations as the transmission line. Since copper has a lower sheet resistance than aluminum, the use of copper for the integrated transmission line can reduce its resistance even more. Of course, other conductive materials could be used. Furthermore, embodiments of the present invention may use very high speed transistor devices. These fast devices allow the designer to create on-chip high-speed components included in a frequency generator or voltage controlled delay, for example, using an integrated transmission line as a timing reference device. [0121] Embodiments of the present invention include a fully integrated frequency generator that uses a feedback loop to force a voltage controlled oscillator (VCO) to have a period equal to the time delay of an integrated transmission line. The integrated transmission line (ITL) resides on the same monolithic silicon die along with other circuitry contained within each of the functional blocks of the frequency generator. Embodiments of the invention include a frequency generator that is totally integrated and uses no external devices such as a quartz crystal as a reference device. The integrated transmission line may be used as the timing reference device for the frequency generator. The frequency generator may be used to generate a digital system clock, for example.
[0122] Fig. 3 illustrates a frequency generator according to one embodiment of the present invention. Frequency generator 300 includes a feed forward path of a feedback loop comprising a phase detector ("PD") 310, loop filter 320 and voltage controlled oscillator ("VCO") 330 coupled in series. PD 310 may detect the phase difference of two input signals REF and FB. In some specific embodiments, PD 3 IO may include a charge pump (not shown) for translating phase differences between REF and FB into currents that are received by loop filter 320 and translated into a voltage. In one embodiment, a phase detector used in a feedback loop may be implemented using the techniques described below and in U.S. Patent Application No. 11/200,472 filed on August 8, 2004, entitled "Circuits and Methods for Reducing Static Phase Offset using Commutating Phase Detectors," naming Paul William Ronald Self as Inventor, the entire disclosure of which is hereby incorporated herein by reference. The output of the PD 310 is coupled to the input of VCO 330 through loop filter 320, which may be a resistor and capacitor connected in series, for example. VCO 330 generates a signal, VOUT, having a frequency, FOUT, that is dependent on the voltage received from filter 320. The output of the VCO may be the output of the system, which may be used
as a reference signal sxich as a system clock, or which may be used to generate yet other signals at other frequencies.
[0123] The output of the VCO is also provided through a feedback path of the feedback loop to the two inputs of PD 310. The feedback path includes an integrated transmission line 350 for introducing a delay that acts as the timing reference for the frequency generator. The delay caused by the integrated transmission line is referred to herein synonymously as "time delay" and "flight time delay." For instance, the signal at the VCO output may be delayed by a predetermined amount of time as it passes through integrated transmission line 350. The delayed VCO output signal is then received at one input of PD 310. An undelayed version of the VCO output signal is provided to the other input of PD 310. Thus, PD 310 receives both delayed and undelayed versions of the VCO output through the feedback path. The action of the loop will cause the period of the VCO output signal to be the same as the delay caused by integrated transmission line 350.
[0124] Embodiments of the present invention may also include lock control circuit 340. To achieve desired performance, frequency generator 300 should operate so that the period of the VCO output signal, Tvco (i.e., 1/ Foim where Fout is the frequency of the VCO output signal), is equal to the value of the integrated transmission line's flight time delay, τχL. Some embodiments of the invention may include a lock control circuit 340 to help eliminate any ambiguity as to what value the period of V<χjτ will be locked. Lock control circuit 340 acts as a lock acquisition aid. for the system so that the VCO output signal has a period equal to the time delay of the integrated transmission line. For example, lock control circuit 340 may ensure that the phase detector is locking to appropriate edges of the feedback signals REF and FB. Without lock control circuit 340, Tvco could lock to a value that is equal to TTL divided by an integer. That is, instead of locking to a value equal to τTL, Tvco could lock to a value equal to ττi/2 or iχi/3 or TTL/4 etc. Lock control circuit 340 guarantees that PD 310 will only compare two successive rising edges of VOUT- Without lock control circuit 340, the loop could potentially enter a locked condition where PD 310 is comparing two non-successive rising edges of VOUT- If the loop enters a stable state where PD 310 is comparing non- successive edges from the VCO, this would mean that Tvco would be locked to a value equal to ττi/2 or τχi/3 or tτi/4 etc. By introducing lock control circuit 340 in the loop, only successive rising edges "will be compared by PD 310, and Tyco will be forced to only equal TTL. In this example, lock control circuit 340 is between VCO 330 and integrated transmission line 350. However, lock control circuit 340 may be located at a variety of
positions in the circuit and may further include connections to control the operation of other loop blocks.
[0125] Fig. 4 illustrates a frequency generator according to another embodiment of the present invention. Frequency generator 400 includes a voltage controlled oscillator ("VCO") 430, a phase detector 410, a charge pump 420, an integrating capacitor ("Ci") 480 and zero resistor ("Ri") 481, pulse removal logic 440, integrated transmission line 450 and an enable control circuit including AND gate 460 and D-type flip flop ("DFF") 461. Integrated transmission line 450 introduces a delay in the feedback path for controlling the frequency of the VCO output signal. Pulse removal logic 440 and enable control circuits 460-461 are the lock control circuits for this example. These circuits eliminate ambiguity as to what value the period of V"ouτ will be locked to. The operation of frequency generator 400 may be referred to as either phase locked loop ("PLL") operation, because it includes a phase detector and VCO, or as delay locked loop ("DLL") operation because it forces the period of the voltage controlled oscillator to be equal to the flight time delay of the integrated transmission line. [0126] Fig. 5 is an example of a phase detector according to one embodiment of the present invention. Phase detector 500 includes two D-type flip flops 501 and 502 ("DFF"), AND gate 503, inverter 504 and OR gate 505. In this example, AND gate 503, inverter 504 and OR gate 505 provide both enable and self-reset functions for phase detector 500. DFF 501 is initialized into a logic low state (e.g., Vss or ground) and receives signal REF as a clock input, which, in frequency generator 200 is the output of the integrated transmission line 250 (i.e., the delayed VCO output signal). The D-input is connected to power supply Vdd. Thus, when the phase detector receives the delayed signal, the Q-output switches from low to high. Similarly, DFF 502 receives signal VCO as a clock input, which in frequency generator 200 is the output of the pulse removal logic 240 (i.e., the undelayed VCO output signal). The D- input of DFF 502 is also connected to power supply Vdd. Thus, when the phase detector receives the undelayed signal, the Q-output switches from low to high. Regardless of whether REF or VCO is received first, the circuit is reset to low logic state when both outputs are high using AND gate 503, which has inputs connected to the Q-outputs of DFFs 501 and 502 and an output coupled to the reset terminal of each DFF 501 and 502 through OR gate 505. Phase detector 500 also includes an ENABLE input. When the ENABLE input is in a high logic state (e.g., Vdd), the output of inverter 504 is logic low and phase detector 500 is enabled. However, when the ENABLE input is in a low logic state, then the output of
inverter 504 Is high and phase detector 500 is disabled because the reset is held high. A high logic level for reset forces the Q outputs of the DFFs 501 and 502 to logic low levels.
[0127] The operation of frequency generator 400 using the example phase detector 500 can be understood by referring to timing diagrams found in Figs. 6-8. Fig. 6 is an example waveform to illustrate pulse removal according to one embodiment of the present invention. The timing diagram of Fig. 6 shows the operation of the pulse removal logic ("PRL") 440. PRL 440 removes pulses from the signal VOuτ, which is the output of the VCO 430. In the example shown in Fig. 6, VPRL is generated by deleting 75% of the Vouτ pulses. Depending on the requirements for the loop, a higher percentage or a lower percentage of the pulses could be removed. In one embodiment, at least a majority of the pulses are removed. The percentage of" pulses removed may range from 50% to a little less than 100%. PRL 440, together with the enable control circuits 460 and 461 and the ENABLE function of the phase detector ("PD") 410 are included in the loop in order to eliminate any ambiguity as to what value the period of VOUT will be locked to. To achieve desired performance, frequency generator 400 should operate so that the period of the VCO output signal, Tvco, is equal to the value of Hie integrated transmission line's flight time delay, TTL- Without pulse removal logic 440, Tvco could lock to a value that is equal to τTL divided by an integer. That is, instead of locking to a value equal to τTL, Tyco could lock to a value equal to τχi/2 or τTi/3 or TTL/4, etc. PRX 440 and the ENABLE input of PD 410 guarantee that PD 410 will only compare two successive rising edges of VOUT- Without PRL 440 and the PD ENABLE, the loop could potentially enter a locked condition where PD 410 is comparing two non- successive rising edges of VOUT- If the loop enters a stable state where PD 410 is comparing non-successive edges from the VCO, this would mean that Tvco would be locked to a value equal to τχi/2 or τTi/3 or τχi/4, etc. Thus, some embodiments of the present invention may include PRL 440 and an ENABLE function in PD 410, or other lock control circuitry, so that Tyco will be forced to equal TTL-
[0128] Fig. 7 is a timing diagram to illustrate the operation of a frequency generator according to one embodiment of the present invention. The timing diagram found in Fig. 7 illustrates the behavior of the circuit when Tvco is too long (i.e., FOUT is too low). As is shown in Fig. 7 the output of PRL 440, VPRL, contains pulses that have the same pulse width as those of its input, which is the output of VCO 430, VOUT- However, the number of pulses in VPRL is reduced from the number found in VOUT- VTL is a delayed version of VPRL with a time delay equal to τTL. Initially, VPRL, VTL, VEN, UP and DOWN are all logic low. With Vw
being low, the ENABLE function of the PD is de-asserted which forces the PD outputs, UP and DOWN, to logic low states. The first rising edge of VPRL clocks the D-flip flop QEN 461, which results in VEN going to a logic high state. When VEN goes high, PD ENABLE becomes asserted which means that the outputs of the PD are allowed to change states as a function of the PD inputs, VCO and REF. Because of the propagation delay of the D-flip flop QEN, the PD ENABLE is not asserted for the first rising edge of VPRL, and therefore, the PD does not respond to the first rising edge of VPRL. In this example, the period of VOUT, TVCO, and thus the period of VPRL, is longer than TTL- Since TTL is shorter than Tvco, the first rising edge that the PD responds to comes from VTL- The rising edge of VTL causes the UP output of the PD to go to a logic high. An UP signal going high before the DOWN signal goes high indicates that the frequency of the VCO needs to increase (i.e., Tyco needs to decrease). A high logic level for the UP input of the charge pump ("CP") 420 causes the CP to turn on a charging current source (not shown), which charges the capacitor ("Cj") 480. The CP charging current into Ci results in a voltage ramp in the positive direction for Vi, thereby increasing the frequency of Vout (i.e., decreasing the period Tyco)- The decrease in Tvco is shown in the timing diagram of Fig. 7. The end result is that the feedback loop causes Tvco to move in a direction to bring it closer in value to TTL3 which is the time delay of the integrated transmission line.
[0129] When VPRL goes high the second time, the PD responds by forcing the DOWTSf output signal high. For a short amount of time both PD outputs, UP and DOWN, are at logic highs simultaneously. When DOWN goes to a logic high, the CP will turn on a discharging current source (not shown). Thus, while the UP and DOWN signals are both high, current sources in the charge pump are charging and discharging capacitor ("Ci") 480 simultaneously. In one embodiment the charge pump current sources are equal in value, and therefore, when the DOWN signal goes high the charging of Ci ends and the positive voltage ramp of Vi stops. When both PD outputs (i.e., UP and DOWN) are logic high, DFFs 501 and 502 in the PD are reset through AND gate 503 and OR gate 505, causing both PD outputs to go logic low simultaneously. When both PD outputs are logic high, AND gate 460 is also activated. Thus, AND gate 460 generates a signal that resets D-flip flop QEN, which results in VEN going to a logic low.
[0130] Fig. 8 is another timing diagram to illustrate the operation of a frequency generator according to one embodiment of the present invention. The timing diagram of Fig. 8 illustrates the behavior of the invention when Tyco is too short (Le. the FOUT is too high).
Initially, VPRL, VTL, VEN, UP and DOWN are all logic low. The first rising edge of VPRL clocks D-flip flop QEN which results in VEN going to a logic high. When VEN goes high, PD ENABLE becomes asserted, which means that the outputs of the PD are allowed to change states as a function of the PD inputs, VCO and REF. As was previously explained, because of the propagation delay of the D-flip flop QEN, the PD ENABLE is not asserted for the first rising edge of VPRL. Therefore, the PD does not respond to the first rising edge of VPRL- Since Tyco is shorter than XTL, the PD responds first to the second rising edge of VPRL. When the second rising edge of VPRL occurs, the DOWN output of the PD goes to a logic high. A DOWN signal going high before the UP signal goes high indicates that the frequency of the VCO needs to decrease (i.e . , Tyco needs to increase). A high logic level for the DOWN input of CP 420 causes the CP to turn on a discharge current source. The discharging current source results in charge being removed from Cj and also results in a negative going voltage ramp for V1. When the first rising edge of VTL occurs, the PD responds by forcing the UP output to a high logic level. When the UP input of the CP goes to a high logic level, the CP turns on a charging current source. At this point both the charging and discharging current sources on the output of the CP are turned on. Because the current sources are of equal value, the discharge of Ci ends and. the negative going voltage ramp of Vi stops. For a short amount of time both the UP signal and DOWN signal are at logic highs simultaneously, and then the PD resets itself which results in both PD outputs (UP and DOWN) going to logic lows simultaneously. At the same time that the PD resets itself, AND gate 460 generates a signal that resets D-flip flop QEN, causing VEN to go to a logic low. A low VEN de-asserts the ENABLE function of the PD. The lower voltage of Vi causes the period of the VCO to increase (i.e., the frequency of the VCO decreases). This increase in Tvco is shown in the timing diagram of Fig. 8. Again, the result is that the feedback loop has caused Tvco to move in a direction to bring it closer in value to TTL.
[0131] hi the manner described in the previous paragraphs, the loop will gradually adjust Vi and the period of the VCO until the value of Tvco matches the flight time delay of the integrated transmission line. Thus, one can see that an integrated transmission line can. be used as a timing reference for a completely integrated oscillator rather than using an external device such as a quartz crystal as a reference.
[0132] Frequency generators may be implemented in other ways using an integrated circuit feedback loop with an integrated transmission line. Fig. 9 illustrates a frequency generator 900 according to another embodiment of the present invention. As shown in Fig. 9, a voltage
controlled oscillator (VCO) can be designed to have more than one output phase. For example, a VCO may be designed to have N equally spaced output phases (e.g., φi, Φ2, Φ3, ... s ΦN, where N is an integer representing the Nth VCO output phase). In this embodiment, the frequency of the VCO may be set by using the feedback loop to control the time difference (i.e., phase difference) between any two of the VCO>'s output phases. The relationship between the frequency of the VCO and the time difference between two adjacent output phases is shown below.
Fvco = 1/Tvco = l/NτD
Where, Fyco is the frequency of the VCO, Tyco is tb.e period of the VCO3 τD is the magnitude of the time difference between two adjacent VCO output phases and N is the number of equally spaced VCO output phases.
[0133] In the example circuit shown in Fig. 9, the time difference between two adjacent output phases (e.g., φi, Φ2) are controlled by the loop. However, it is to be understood that it is possible to set the frequency of the VCO by controlling the time difference between any two output phases (e.g., ΦL, ΦM, where L and M are any two of the VCO output phases). Thus, Fig. 9 is a frequency generator that uses a feedback loop to force two adjacent output phases of a VCO to be equal to the flight time delay of an integrated transmission line 950. Frequency generator 900 includes a phase detector 9 10, charge pump 920, loop filter (e.g., capacitor 980 and resistor 981), VCO 930 and integrated transmission line 950 configured in a feedback loop. In this example, the circuit further includes a lock control circuit comprising two pulse removal logic ("PRL") circuits 941 and 942. Therefore, in this example, VCO 930 provides two outputs Voi (e.g., <)> 1) and V02 (e.g., Φ2), which are coupled to the input of each PRL circuit. The input of PRL 9*41 receives Voi (e.g., φO, and the output of PRL 941 (i.e., signal "VPRU") is coupled to the input of integrated transmission line 950 and the clock input of D flip-flop 961. Similarly, the input of PRL 942 receives V02 (e.g., φ2), and the output of PRL 942 (i.e., signal "VpRL2") is compled to the input of phase detector 910.
[0134] Generally, the embodiment in Fig. 9 is another example of an integrated circuit with a feedback loop that contains an integrated transmission line that is used as a timing reference. The integrated transmission line resides oxi the same monolithic silicon die along with all the circuitry contained within each of the functional blocks shown in Fig. 9. One example implementation of the phase detector (PD) was previously shown in Fig. 5. An example operation of the Pulse Removal Logic (PRl-) was previously shown in Fig. 6. Each
PRL 941 and 942 removes pulses from the corresponding VCO output phase to which it is connected. The purpose and operation of a PRL block was previously explained.
[0135] The timing diagrams in Figs. 10 and 11 illustrate how the embodiment of Fig. 9 uses a feedback loop to force τo "to equal the flight time delay of the integrated transmission line, TTL. For the examples found in Figs. 10 and 11, the VCO is assumed to have 4 equally spaced output phases (N = 4) producing signals V01, V02, V03 and V04. However, it should be noted that the same principle of operation applies for any number, N, of VCO output phases.
[0136] Fig. 10 is a timing diagram to illustrate the operation of the frequency generator of Fig. 9 according to one embodiment of the present invention. The timing diagram found in Fig. 10 illustrates the behavior of the circuit when τo and Tyco are too long (i.e., FOUT is too low). As is shown in Fig. IO the outputs of the PRL 941 and 942, VPRLI and VPRL2, contain pulses that have the same pulse width as those of their inputs, Voi and V02. However, the number of pulses in VPRLi and VPRL2 is reduced from the number found in Voi and Υo2. VTL is a delayed version of VPRLI with a time delay equal to flight time delay of the integrated transmission line, TTL. Initially, VPRLI, VPRLZ, VTL, VEN, UP and DOWN are all logic low. With VEN being low, the EIsTABLE function of the PD is de-asserted, which forces the PD outputs, UP and DOWN, to logic low states. The first rising edge of VPRLI clocks the D-flop QEN, which results in VEN going to a logic high state. When VEN goes high, PD ENABLE becomes asserted, which means that the outputs of the PD are allowed to change states as a function of the PD inputs, VCO and REF. Since τD is longer than τTL, the first rising edge that the PD responds to comes from VχL. The rising edge of VTL causes the UP output of the PD to go to a logic high. An UP signal going high before the DOWN signal goes high indicates that the frequency of the VCO needs to increase (i.e., τD and Tyco need to decrease). A high logic level for the UP input of the Charge Pump (CP) causes the CP to turn on a charging current source (not shown) connected to its output. The current source charges the capacitor 980 ("Ci") which results in a voltage ramp in the positive direction for Vi . When VPRL2 goes high, the PD responds by forcing the DOWN output signal high. For a short amount of time both PD outputs, UP and DOWN, are at logic highs simultaneously. When DOWN goes to a logic high, the CP will turn on a discharging current source (not shown) that is connected to its output. While the UP and DOWN signals are both high, the Charge Pump's charging and discharging current sources are on simultaneously. In one embodiment, the charging and discharging current sources are equal in value, and thus, the charging of Ci ends and the positive voltage ramp of Vi stops. At this point the PD resets itself which
results in both PD outputs, UP and DOWN, going to logic lows simultaneously. At the same time that the PD resets itself, AND gate 960 ("Gl") generates a signal that resets D-flop 961 ("QEN")S which results in VEN going to a logic low. A low VEN de-asserts the ENABLE function of the PD. The higher voltage OfV1 causes τo and Tyco to decrease (i.e., the frequency of the VCO increases). This decrease in τD and Tvco is shown in the timing diagram of Fig. 9. The end result is that the feedback loop of the invention has caused τø to move in a direction to bring it closer in value to the flight time delay of the integrated transmission line, XTL.
[0137] Fig. 11 is another timing diagram to illustrate the operation of the frequency generator of Fig. 9 according to one embodiment of the present invention. The timing diagram of Fig. 11 illustrates the behavior of the circuit when τo and Tyco are too short (i.e. the Four is too high). Just like the previous example initially, VPRLI, VPRL2, VTL, VEN, UP and DOWN are all logic low. The first rising edge of VPRLI clocks D-flop QEN, which results in VEN going to a logic high. When VEN goes high, PD ENABLE becomes asserted, which means that the outputs of the PD are allowed to change states as a function of the PD inputs, VCO and REF. Since to is shorter than TTL, the PD responds first to the rising edge of VPRL2- When the rising edge of VPRL2 occurs, the DOWN output of the PD goes to a logic high. A DOWN signal going high, before the UP signal goes high indicates that the frequency of the VCO needs to decrease (i.e., XD and Tyco need to increase). A high logic level for the DOWN input of the Charge Pump (CP) causes the CP to turn on a discharge current source that is connected to its output. The discharging current source results in charge being removed from Ci and also results in a negative going voltage ramp for Vi. When the first rising edge of VTL occurs, the PD responds by forcing the UP output to a high logic level. When the UP input of the CP goes to a high logic level, the CP turns on a charging current source that is connected to its output. At this point both the charging and discharging current sources on the output of th.e CP are turned on. In one embodiment, the charging and discharging current sources are equal in value, and thus, the discharge of Ci ends and the negative going voltage ramp of Vi stops. For a short amount of time both the UP signal and DOWN signal are at logic highs simultaneously and then the PD resets itself which results in both PD outputs, UP and DOWN, going to logic lows simultaneously. At the same time that the PD resets itself, AND Gate Gl generates a signal which resets D-flop QEN causing VEN to go to a logic low. A low VEN de-asserts the ENABLE function of the PD. The lower voltage of Vi causes %u and Tyco to increase (i.e., the frequency of the VCO decreases). This
increase in TD and Tvco is shown in the timing diagram of Fig. 11. Again, the result is that the feedback loop of the invention has caused τo to move in a direction to bring it closer in value to the flight time delay of ttie integrated transmission line.
[0138] In the manner described in the previous two paragraphs, the feedback loop will gradually adjust Vi and the frequency of the VCO until the value of τo matches the flight time delay of the integrated transmission line.
[0139] Fig. 12A illustrates an integrated transmission line used as a timing reference in a feedback loop according to one embodiment of the present invention. In one embodiment, the present invention includes an integrated circuit comprising a first controlled delay line 1220 coupled in a feedback loop and an transmission line 1210, where the transmission line is used as a timing reference for the feedback loop. In one embodiment, the feedback loop and the transmission line are integrated on a single integrated circuit. The feedback loop may further include a phase detector 1230 and loop filter 1240. Phase detector 1230 includes a first input coupled to an output of transmission line 1210 and a second input coupled to an output of controlled delay line 1220. The output of phase detector 1230 is coupled to loop filter 1240. The loop filter may include a capacitor, for example. The output of loop filter 140 is used to control the delay of controlled delay line 1220.
[0140] In one embodiment, a reference signal generator 1201 may be used to generate a reference signal, which may be a periodic signal such as a square wave, for example. The reference signal is provided to the inputs of transmission line 1210 and controlled delay line 1220. Transmission line 1210 delays the reference signal by a certain delay dependent on the characteristics of the transmission line. The controlled delay line 120 delays the reference signal by a delay that is a function of the signal received on the control input ("ctrl"). Phase detector 1230 receives a first delayed signal from the transmission line and a second delayed signal from the controlled delay line. Phase detector 1230 may be used to detect a phase difference between the two delayed signals. The output of phase detector 123O may be filtered in loop filter 1240, and the output of loop filter 1240, which may represent the difference in delay between transmission line 1210 and controlled delay line 1220, may be coupled to the control input of controlled delay line 1220. Thus, the loop increases the delay of the controlled delay line if the transmission line delay is greater than the delay of the controlled delay line. Similarly, the loop decreases the delay of the controlled delay line if the transmission line delay is less th.an the delay of the controlled delay line. The action of
the feedback loop will cause the delay of the controlled delay line 122O to be equal to the delay of the transmission line 1210. The control signal generated by trxe loop filter, which is used to control the delay line 1220, may also be used to control the delay of one or more other delay lines 1250. The other delay lines may receive input signals, such as digital data signals or system clocks, for example, and delay the input signals by an. amount of time that is related to the delay of controlled delay line 1220.
[0141] Fig. 12B illustrates a controlled delay line circuit with an integrated transmission line reference according to one embodiment of the present invention. Ln one embodiment, the present invention includes an integrated circuit that uses a feedback loop to force a controlled delay line to have a time delay that is equal to the flight time delay of tbxe integrated transmission line (ITL). While the example in Fig. 12B is described using a voltage controlled delay line, it is to be understood that a variety of different implementations of a controlled delay line may be used, including current controlled delay lio.es and digital controlled delay lines (i.e., a digital delay line, such as series connected inverters and a multiplexer, for example).
[0142] Controlled delay line circuit 1200B includes an RC oscillator 1261 coupled to a reference voltage controlled delay line 1263 and an integrated transmission line 1262. The reference voltage controlled delay line 1263 is included in a feedback loop. In this example, the feedback loop also includes a phase detector 1264, a charge pump 1265, and a capacitor 1266 ("C1"). The voltage on Ci (i.e., Vi) is used to control both the refer-ence voltage controlled delay line 1263 and one or more other voltage controlled delay lines 1267. A time difference between a first delayed signal from the transmission line 1262 and a second delayed signal from the controlled delay line 1263 may be translated into a voltage on a capacitor and used to control the first variable delay of delay line 1263. The other voltage controlled delay line 1267 may receive data and produce delayed data in. accordance control voltage Vi. Fig. 13 is an example of a phase detector according to one embodiment of the present invention. The phase detector 1300 in Fig. 13 is a phase-frequency detector including two D-rype Flip Flops ("DFF") 1310 and 1311, a reset circuit including an AND gate 1312, and an enable circuit including an inverter 1313. The reset and enable circuits are combined using aNOR gate 1314 that drives the reset inputs of the DFFs. The phase detector of Fig. 13 is one possible way to implement the phase detector of Fig. 12B. As xnentioned above, a phase detector may be implemented using the techniques described below and in U.S. Patent Application No. 11/200,472 filed on August 8, 2004, entitled "Circuits and Methods for
Reducing Static Phase Offset using Commutating Phase Detectors," naming Paul William Ronald Self as Inventor, the entire disclosure of which is hereby incorporated herein by reference.
[0143] The integrated transmission line resides on the same monolithic silicon die along with all the circuitry contained within each of the functional blocks shown in Fig. 12B.
Embodiments of the present invention include an implementation of a totally integrated delay locked loop that does not use the period of a system clock as the timing reference, but rather, an integrated transmission line is used as the timing reference device.
[0144] The present embodiment includes a variety of advantageous features for the delay locked loop that may be included to improve the performance of the DLL. Before describing the operation of Fig. 12B, such features will be described, hi this example, a periodic reference signal is generated by RC oscillator 1261 and provided to the inputs of the reference voltage controlled delay line 1263 and an integrated transmission line 1262. hi other embodiments, a periodic signal may be generated by other circuits on the same integrated circuit or external to the integrated circuit and coupled to inputs of the transmission line and controlled delay line. During each period of the output signal of the RC oscillator (RCO), the amount of time that the RCO output signal is at a high logic level is larger than the flight time delay of the integrated transmission line. Further, the amount of time that the RCO output signal is at a high logic level is also larger than the longest possible time delay of the VCDL. An ENABLE function for the phase detector (PD) is another feature, hi this embodiment, the PD includes an ENABLE function in order to guarantee that the phase detector correctly interprets the phase information from its inputs, INl and IN2. It is theoretically possible for the PD to incorrectly interpret the phase information at its inputs and then to continue to incorrectly interpret the phase relationship on subsequent cycles of the RC oscillator (RCO). By adding an ENABLE function to the PD, the PD is reset on every cycle by the RCO output signal, and thus, even if the PD incorrectly interprets the input phase relationship once, the ENABLE function will prevent it from continuing to incorrectly interpret the input phase relationship on subsequent cycles of the RCO.
[0145] The operation of the invention can be understood by referring to timing diagrams found in Fig. 14 and Fig. 15. The timing diagram found in Fig. 14 illustrates the behavior of the circuit when the time delay of the VCDL, TDL, is smaller than the time delay of the integrated transmission line (ITL), TTL- VDL is a delayed version of VRCO with a time delay
equal to TDL- VTL is also a delayed version of VRCO with a time delay equal to τn- Initially, VRCO, VTL) VDL, UP, and DOWNT are all logic low. With VRCO being low, the ENABLE function of the PD is de-asserted which forces the PD outputs, UP and DOWN, to logic low states. The first rising edge of VRCO launches a rising edge into the two delay lines, the ITL and the VCDL. When VRCO goes high, PD ENABLE becomes asserted which means that the outputs of the PD are allowed to change states as a function of the PD inputs, INl and IN2. Since TDL is shorter than TTL, the first rising edge that the PD responds to comes from VDL- The rising edge of VDL causes the UP output of the PD to go to a logic high level. An UP signal going high before the DOWN signal goes high indicates that the time delay of the VCDL needs to increase. A high, logic level for the UP input of the charge pump (CP) causes the CP to turn on a charging current source connected to its output. The current source charges the capacitor Ci which results in a voltage ramp in the positive direction for Vi. When VTL goes to a high logic level, the PD responds by forcing the DOWN output signal high. For a short amount of time t>oth PD outputs, UP and DOWN, are at logic high levels simultaneously. When DOWN goes to a logic high, the CP will turn on a discharging current source that is connected to its output. While the UP and DOWN signals are both high, the CP's charging and discharging current sources are on simultaneously. Because the current sources are equal in value, the charging of Ci ends and the positive voltage ramp of Vi stops. At this point the PD resets itself which results in both PD outputs, UP and DOWN, going to logic low levels simultaneously. At some point after the PD resets itself, VRCO goes to a logic low which de-asserts the ENABLE function of the PD. The higher voltage of Vi causes the time delay of the VCDL to increase. This increase in τDiJs shown in the timing diagram, of Fig. 14. The end result is that the feedback loop has caused τoLto move in a direction to bring it closer in value to TTL- [0146] The timing diagram found in Fig. 15 illustrates the behavior of the circuit when the time delay of the VCDL, τDL, is longer than the time delay of the integrated transmission line (ITL), TTL- Just like the previous example initially, VRCO, VTL, VDL, UP, and DOWN are all logic low. With VRCO being low, the ENABLE function of the PD is de-asserted which forces the PD outputs, UP and DOΛVN, to logic low states. The first rising edge of VRCO launches a rising edge into the two delay lines, the ITL and the VCDL. When VRCO goes high, PD ENABLE becomes asserted which means that the outputs of the PD are allowed to change states as a function of the PD inputs, INl and IN2. Since TDL is longer than TTL, the first rising edge that the PD responds to comes from VTL. The rising edge of VTL causes the
DOWN output of the PD to go to a logic high. A DOWN signal going high before the UP signal goes high indicates that the time delay of the VCDL needs to decrease. A high logic level for the DOWN input of the charge pump (CP) causes the CP to turn on a discharging current source connected to its output. The current source discharges the capacitor Ci which results in a voltage ramp in the negative direction for Vi. When VDL goes to a high logic level, the PD responds by forcing the UP output signal high. For a short amount of time both PD outputs, UP and DOWN, are at logic high levels simultaneously. When UP goes to a logic high, the CP will turn on a charging current source that is connected to its output. While the UP and DOWN signals are both high, the CP's charging and discharging current sources are on simultaneously. Because the current sources are equal in value, the discharging of Ci ends and the negative voltage ramp of Vi stops. At this point the PD resets itself which results in both PD outputs, UP and DOWN, going to logic low levels simultaneously. At some point after the PD resets itself, VRCO goes to a logic low which de- asserts the ENABLE function of the PD. The lower voltage of Vi causes the time delay of the VCDL to decrease. This decrease in TDL ΪS shown in the timing diagram of Fig. 15. Again, the end result is th.at the feedback loop of the invention has caused TDL to move in a direction to bring it closer in value to τTL.
[0147] In the manner described in the previous two paragraphs, the feedback loop will gradually adjust Vi and the time delay of the voltage controlled delay line until the value of TDL matches the flight time delay of the integrated transmission line. The examples illustrated in Figs. 14-15 are commonly referred to as delay locked loops. Thus, one can see that an integrated transmission line can be used as a timing reference device for a completely integrated feedback loop.
[0148] Described next are techniques for reducing static phase offset in timing loops. In the following description, for purposes of explanation, numerous examples and specific details are set forth in order to provide a thorough understanding of embodiments of the present invention. It will be evident, however, to one skilled in the art that the embodiments of the invention as defined by the claims may include some or all of the features in these examples alone or in combination with other features described below, and may further include obvious modifications and equivalents of the features and concepts described herein.
[0149] The present disclosure further includes circuits and methods that may be used to implement phase detection. The following circuits and methods may be used in a variety of
applications that may benefit from improved phase detection. The following circuits and methods may also be used to improve the accuracy of the frequency generation and delay controls disclosed above. For example, the phase detectors disclosed above may be implemented using the techniques that will now be described. [0150] Embodiments of the present invention reduce static phase offset by detecting mismatches in the phase detector, such as mismatched delays or currents, that contribute to static phase offset, and calibrate out the mismatches to reduce static phase offset. In one embodiment, a timing loop includes first and second phase detectors, wherein during a first time period, the first phase detector is coupled in a closed timing loop and the second phase detector is decoupled from the closed timing loop and calibrated. During a second time period, the second phase detector is coupled in a closed timing loop and the first phase detector is decoupled from the closed timing loop and calibrated. By commutating between phase detectors, one phase detector may be used to control the timing loop while the other phase detector is calibrated. After one phase detector is calibrated, it may be used to control the loop while the other phase detector is calibrated. Calibration may be used to eliminate or reduce mismatches in each phase detector that contribute to static phase offset error. In one embodiment, the phase detectors are alternately configured in a correction loop to perform calibration. A correction loop may detect operational parameters of the phase detector that relate to static phase error and generate a correction signal that is fed back into the phase detector being calibrated to null out (or autozero) the error. For example, a correction loop may detect mismatches, such as mismatched currents, in the phase detector. When phase detector is in a closed timing loop, these mismatches may result in a static phase offset. Thus, the correction signal may adjust the operation of the phase detector to reduce the mismatch, and thereby reduce the static phase offset. In one embodiment, mismatch may result in a voltage drift that is indicative of one or more mismatches in the phase detector.
For example, the voltage drift may be at the output of trie phase detector. Ideally, if both inputs of the phase detector receive the same input, then there should be no voltage drift at the output. Therefore, in one embodiment, a common input signal may be applied to both inputs of the phase detector being calibrated. The correction loop may detect any voltage drifts that may arise while the phase detector is receiving a common input, and generate a correction signal to reduce the mismatch causing the dribft.
[0151] Fig. 16A illustrates a phase locked loop according to one embodiment of the present invention. In one embodiment, phase locked loop 160(LA includes first and second phase
detectors 1610A and 1610B that are alternately coupled to an mismatch detector 1660 and a voltage controlled oscillator 1630 during different time periods. For example, in order to reduce static phase offset in the system, one phase detector may be used to control the timing loop while the other phase detector goes through a calibration cycle (i.e. a correction cycle). In this example, a reference signal ("REF") is received in both phase detectors, and the phase detector outputs are alternately coupled through a loop filter 1620 and VCO 1630 to feedback inputs of each phase detector (here labeled "VCO"). When a correction cycle is completed, the phase detectors alternate (i.e., exchange roles), and the corrected phase detector is used to control the timing loop while the other phase detector goes through a correction cycle. In other words, embodiments of the present invention periodically disconnect one phase detector from the timing loop to correct the static phase offset. However, loop control must be maintained. Thus, two phase detectors are included so that when one phase detector is disconnected from the loop to be corrected, the other phase detector is configured into the loop arid used to maintain control of the loop. Once one phase detector has been corrected, the system is reconfigured to switch the corrected phase detector into the loop and the other phase detector out of the loop for static phase offset correction. Thus, the system alternates (commutates) between two phase detectors to reduce static phase offset while maintaining control of the loop.
[0152] During static phase offset correction, the feedback input (here, VCO) of one of the phase detectors (e.g., PD 1610A) is decoupled from the timing loop (e.g., switch S6 is opened) and coupled to a common input signal (e.g., switch S2 is closed). In this example, the common input signal is the REF input. However, other signals could be used. The output of the phase detector to be calibrated is decoupled from the timing loop (e.g., switch S9 is opened) and into a correction loop. For example, the output of the phase detector may be coupled to mismatch detector 1660 (e.g., switch Sl 1 is closed). Ideally, when the loop is open and the same signal is applied to both inputs of the phase detector, there should be no net change in the output of the phase detector if the delays and circuits are perfectly matched. For example, if the delay between REF and the output is the same as the delay between VCO and the output, or if the circuitry is matched, then there should be no static phase offset. However, typically the delays are different and the circuits are mismatched due to manufacturing variations and imperfections. Thus, if the same signal is applied to both inputs, there will be a net change at the output. The net change 'will translate into a static phase offset between the REF and VCO signals when the timing loop is closed. Accordingly,
embodiments of the present invention detect the change at the output of the phase detector when a common signal is applied to the inputs and adjust the circuitry in the phase detector to reduce the detected change.
[0153] During calibration, mismatch detector 1660 is coupled to the output of the phase detector to detect changes in the phase detector output. Thus, mismatch detector 1660 may include inputs coupled through switches SI l and S12 to alternately detect changes in both phase detectors. The output of mismatch detector 1660 includes a correction signal that is coupled to an adjust input of phase detector 1610A and an adjust input of phase detector 161OB. A variety of circuit techniques may be used to detect mismatch in phase detectors, and a variety of correction signals and adjustment techniques may be used to calibrate the phase detector. In particular, if the change in the output of the phase detector is positive, then the adjustments (e.g., to delays, currents, or voltages) should cause the outpixt to change in the negative direction.. Similarly, if the change in the output of the phase detector is negative, then the adjustments should cause the output to change in the positive direction. [0154] In accordance with this embodiment, during a first time period, phase detector 1610A is configured in a closed loop with voltage controlled oscillator 1630, and phase detector 1610B is in a correction loop with mismatch detector 1660, and in accordance therewith, mismatch detector 1660 reduces the static phase offset generated by phase detector 1610B. Similarly, during a second time period, phase detector 1610B is configured in a closed loop with voltage controlled oscillator 1630, and phase detector 1610A is in a correction loop with mismatch detector 1660, and in accordance therewith, mismatch detector 1660 reduces the static phase offset generated by phase detector 1610A. In one embodiment, phase locked loop 1600A further comprises a phase detector configuration controller 1650. Configuration controller 1650 may be used to reconfigure the phase detectors between closed loop configuration and calibration. As illustrated below, an example configuration controller may include a counter to periodically reconfigure the phase detectors.
[0155] Fig. 16B illustrates a delay locked loop ("DLL") including commutating first and second phase detectors 1610A and 1610B that are alternately configured in a closed timing loop while the other is calibrated. The DLL 1600B further includes a loop filter 1620, voltage controlled delay 1640, mismatch detector 1660, and phase detector configuration controller 1650. Loop filter 1620 is coupled through switches S9 and SlO to the outputs of phase detectors 1610A and 1610B. The output of loop filter 1620 is coupled to one input of
voltage controlled delay 1640, and a second input of voltage controlled delay 1640 is coupled to an input signal ("INPUT"). The output of voltage controlled delay 1540 is coupled through switches S6 and S8 to the feedback inputs ("FB") of phase detectors 1610A and 1610B. The input signal is coupled to the reference inputs ("REF") of pliase detectors 1610A and 1610B. The input is also coupled through switches S2 and S3 to the FB input of phase detectors 1610A and 1610B. In this example, when phase detector 1610-.A is being calibrated, switches S2 and Sl 1 are closed, and switches S6 and S9 are open. Thus, both the REF and FB inputs of phase detector 1610A receive the same signal (e.g., here, the input signal) for calibration. During the time that phase detector 1610A is being calibrated, switches S8 and SlO are closed, and switches S3 and S12 are open. Thus, phase detector 1610B is configured to control the timing loop. When the phase detectors commutate (i.e., exchange roles), switches S2 and SIl are opened, and switches S6 and S9 are closed. Thms, phase detector 1610A is configured to control the timing loop. During this time period, switches S8 and SlO are opened, and switches S3 and S 12 are closed. Accordingly, phase detector 1610B may be calibrated while phase detector 161 OA controls the timing loop.
EXAMPLE IMPLEMENTATION OF A PHASE DETECTOR
[0156] Fig. 17 is an example application of a timing loop 1700 with reduced static phase offset according to one embodiment of the present invention. Fig. 17 SIIOΛVS a typical implementation of a zero delay buffer ("ZDB") using a phase locked loop> ("PLL"). PLL 1700 includes an auto-zero phase-frequency detector/charge pump 1710 ("'PFD-CP"), which is a specific implementation of the commutating phase detector approach described above. PLL 1700 further includes a VCO 1720, current source bias generator 1730, and loop filter including resistor 1701 ("Rz"). integration capacitor 1702 ("Ci"), and capacitor 1703 ("CB"). A first input to the PFD-CP 1710 labeled IREF is coupled to a clock signaJ ("CLKEXT"), which may be an external clock, for receiving a signal VFREF- The second input of the PFD- CP 1710 labeled IVCO is coupled to the output of VCO 1720 labeled OUT to receive output signal VCOUT, which may also be the system output. Bias generator 1730 generates the necessary voltages and currents for biasing circuits in PFD-CP 1710 and possibly other circuits.. [0157] Referring to Fig. IB, for an ideal ZDB, the input signal, IREF, would be exactly in phase with the output signal VOUT- This ideal ZDB, therefore, would have a static phase offset of zero seconds. Embodiments of the present invention may be used to substantially
reduce the amount of static phase offset in a ZDB, for exanxple, caused by various imperfections in the semiconductor process. In particular, non-zero static phase offset typically is caused by manufacturing variations and imperfections in the phase-frequency detector ("PFD") and charge pump ("CP"). If the PFD operated ideally, the propagation delay from the VCO feedback input of the PFD to the output DOWN (e.g., see Fig. IB) would be exactly equal to the propagation delay from the REF input to the output UP. If the CP operated ideally, there would be no net charge produced at the output when the inputs, UP and DOWN, receive signals that have exactly the same pulse width. If the PFD and CP together operated ideally, there would be no net charge delivered from the CP output, OUT, when the PFD input signals, IREF and VOuτ, have exactly tfcie same phase. However, since the actual PFD and CP are non-ideal, if IREF and Vouτare exactly in phase, then the CP will produce some non-zero amount of charge at the output and into the PLL' s loop filter. In a closed loop, the PLL will translate the non-zero charge into a. static phase offset between IREF and VOuτ. [0158] In the example shown in Fig. 17, an auto-zeroing pliase detector technique is used to reduce static phase offset. In this example, the voltage on the integration capacitor 1702 is provided at a VCI input of the improved auto-zero PFD-CP 1710. As illustrated by the example in Fig. 18A-B, this voltage may be used, together with the architecture described above, to reduce static phase offset. [0159] Fig. 18A-B is an example of an auto-zero phase detector 1800 ("AZPD") according to one embodiment of the present invention. AZPD 1800 includes two phase-frequency detector and charge pump combinations ("PFD-CP"). The fixst PFD-CP includes PFD 1810A ("PDl") and CP 1811A ("CP1"). The second PFD-C-P includes PFD 1810B ("PD2") and CP 181 IB ("CP2"). An example phase-frequency detector that may be used is shown in Fig. 20. When one of the PFD-CPs is disconnected from the timing loop for mismatch correction, the other PFD-CP is configured into the timing loop to maintain loop control. After one PFD-CP has been through a correction cycle, the other PFD-CP is disconnected from the loop for correction, and the previously corrected PFI)-CP is configured into the loop to maintain loop control. Thus, the two PFD-CP combinations commutate (i.e., exchange) operational states between a correction (or calibration) state and a feedback control state. Thus, each PFD-CP can be periodically corrected while maintaining feedback control of the timing loop.
[0160] In particular, when PFD 1810B and CP 181 IB are configured in a feedback control state and PFD 1810A and CP 181 IA are configured in a correction state, switches 1821 and 1822 are open and switches 1823 and 1824 are closed. The switches in this example are implemented using parallel N-channel and P-channel MOS transistors. Thus, the reference signal input IREF and feedback signal input IVCO are disconnected jfrom the REF and VCO inputs of PDl. Rather, the IREF input is buffered by two inverters 1 SOl and 1802 and provided to both inputs of PDl through switches 1823 and 1824. Tlrus, both the REF and VCO inputs of PDl receive a common input signal. Furthermore, the output OUT of auto- zero phase detector 1800 is decoupled from the output of CPl by opening switch 1829. Instead, the output of CPl is coupled through switch 1830 to calibration capacitor 1812A ("CAZI")-
[0161] It should be noted that prior to entering a correction cycle, sΛvitch 1831 is closed and switch 1830 is open. In this configuration, calibration capacitor 181ZA will be driven by the output of unity gain buffer 186 IA, which in turn receives an input from VCI - the voltage on the integration capacitor 1702 in the loop filter (see Fig. 17). Thus, when the output of CPl is disconnected from the timing loop and connected to calibration capacitor 1812A, the voltage on calibration capacitor 1812A is initially VCI. Additionally;, when the correction cycle begins, switch 1831 is opened so that the voltage on calibration capacitor 1812A may be driven by the output of CP 1. [0162] As mentioned above, at the beginning of a correction cycle, "the voltage on capacitor 1812A is VCI. When switch 1830 is closed, the output of CPl is coupled to the calibration capacitor 1812A. Calibration capacitor 1812A is, in this example, coupled to a window comparator for detecting when the voltage on capacitor 1812A is above or below a threshold. In this example, the threshold of the comparator is generated around VCI using unity gain buffer 1861B ("OP2") and series connected resistors 1862A-B and tøuαsistors 1863A-B. The reason for this is that the static phase offset error in the charge pump may be a function of the voltage on VIN (see Fig. 17). Thus, when a correction cycle occurs it may be desirable to have the voltage on the calibration capacitor to be the same as the voltage at the output of the CP when the CP is configured in the timing loop. However, when PD 1 and CPl are in closed loop mode and the timing loop is locked, there is very little current going into Rz, so the voltage on Ci and CB are substantially the same. Normally, Ci is larger (e.g., 10 times larger) than CB. Thus, in lock VCI and VIN are substantially the same. However, VCI has less electrical noise. Since VIN and VCI are about the same, Vopi is generated from VCI and is
used to set the initial voltage on the calibration capacitor at the beginning of each correction cycle. As mentioned above, VCI is couple through unity gain buffer 186 IA ("OPl") to the calibration capacitor when the PD is in the timing loop. When the PFD-CP enters a correction cycle, Vopi is disconnected from the calibration capacitor so that the voltage on the calibration capacitor can change. In this example, the circuit detects mismatch by determining the polarity of the voltage drift on the calibration capacitor. Thus, when the CP is disconnected from the loop, the calibration capacitor voltage is at the same voltage as the VCI, which is also the center of the comparator window.
[0163] In this example, calibration capacitor 1812A is coupled to a window comparator. Here, capacitor 1812A is coupled to the positive input of a first comparator 1864A and the negative input of a second comparator 1865 A. The negative input of comparator 1854A is coupled to resistor 1862A ("RRFI")> which is the high threshold voltage (i.e., upper trip point), VRFI, of the comparator window. Thus, the voltage on calibration capacitor 1812A, VCAZU will initially be VCI, which is below the high threshold voltage, and the output of comparator 1864A is initially low (i.e., because the negative input, VRFI, is initially at a higher voltage than the positive input, VCAZI). However, if the voltage on the calibration capacitor increases above the high threshold voltage during the correction cycle (e.g., because of circuit mismatches in the PFD or CP), the output of comparator 1864A will transition from low to high. A positive going voltage on the calibration capacitor may indicate that the charge pump is sourcing more current than it is sinking, for example. Thus, to correct this mismatcli, when the window comparator detects a positive going voltage on the calibration capacitor, a. correction signal is generated to trigger a reduction in the charge pump current source and/or increase the charge pump current sink. In this example, the correction signal is a voltage on capacitor 1875 A5 which is coupled to the adjust input of CP 181 IA. The correction signal is generated by coupling the output of comparator 1864A directly to one input of a NANTD gate 1868A and to the other input of NAND gate 1868A through a delay circuit 1866A comprising series connected inverters to create a pulse. The pulse turns PMOS transistor 187 IA on to generate a correction current through resistor 1872A into capacitor 1875A. to change the voltage on the adjust input of CP 181 IA. The correction signal may be used to adjust the source current in CP 181 IA, for example.
[0164] Similarly, calibration capacitor 1812A is coupled to the negative input of a second comparator 1865A. The positive input of comparator 1865A is coupled to resistor 1862B ("RRF2"), which is the low threshold voltage (i.e., lower trip point), VRF2, of the comparator
window. Thus, the voltage on calibration capacitor 1812A, VCAZI, will initially be VCI., which is above the low threshold voltage, and the output of comparator 1865 A is initially low (i.e., because the negative input, VCAZI, is initially at a higher voltage than the positive input, VjRF2). However, if the voltage on the calibration capacitor decreases' below the low threshold voltage during the correction cycle (e.g., because of mismatched circuits in the PFD or CP), the output of comparator 1865A will transition from low to high. A negative going voltage on the calibration capacitor may indicate that the charge pump is sinking more current tfcian it is sourcing, for example. Thus, to correct this mismatch, when the window comparator detects a negative going voltage on the calibration capacitor, a correction signal is generated to trigger a reduction in the charge pump current sink and/or increase the charge pump current source. The correction signal is generated by coupling the output of comparator 1865A directly to one input of a NAND gate 1869A and to the other input of NAND gate 1869A through a delay circuit 1867A comprising series connected inverters to create a pulse. The pulse turns 2STMOS transistor 1874A on to generate a correction current through resistor 1873 A into capacitor 1875A to change the voltage on the adjust input of CP 181 IA.
[0165] As illustrated by the above example, comparators may be used to determine the polarity of the error so that correction may be made in the opposite direction. For noise considerations, a window may be used so that a minimum change in voltage on the calibration capacitor is required before any correction is made. Thus, if a designer opts to use a window comparator implementation to detect mismatch at the output of the phase detector, the voltage range of the comparator window may be set according to the noise tolerance of the system. The amount that the voltage on the calibration capacitor may increase or decrease before triggering a correction signal may be set to achieve target static phase offset tolerance. It is to be understood that other techniques may be used to detect changes at the output of the phase detector, such as by averaging out the noise, for example.
[0166] Fig. 19 is an example of an adjustable charge pump 1900 according to one embodiment of the present invention. Charge pump 1900 ("CP") includes a first input "LTP" coupled through switch 1901 and inverter 1902 to a differential current steering network including PMOS transistors 1903 ("MPC3"), 1904 ("MPC4"), 1910 ("MPC2"), and 1911 ("MPCl"). The gate of transistor 1904 is coupled though inverter 1902 to the "UP" input terminal. When a positive going signal is received on the "UP" terminal, inverter 1902 changes the polarity so that transistor 1904 is activated when an UP signal is received. Ttte UP signal is also coupled through switch 1901 (here, a transmission gate including parallel
PMOS and NMOS devices) to the gate of PMOS transistor 1903. Switch 1901 is used to delay the signal by a time period approximately equal to the delay of inverter 1902. When a positive going signal is received at the UP terminal, transistor 1903 will turn off. Since transistors 1903 and 1904 are biased by common devices 1910 and 1911, a signal received at the UP input terminal will steer current between devices 1903 and 1904. The current through transistor 1904 is used to drive the output of CP 1900. This current is referred to as the charge pump source current.
[0167] Similarly, charge pump 1900 ("CP") includes a second input "DOWN" coupled through switch 1906 and inverter 1907 to a differential current steering network including NMOS transistors 1908 ("MNC3"), 1909 ("MNC4"), 1912 ("MNC2"), and 1913 ("MNCl"). The gate of transistor 1908 is coupled though inverter 1907 to the "DOWN" input terminal. When a positive going signal is received on the "DOWN" terminal, inverter 1907 changes the polarity so that transistor 1908 is inactivated when a "DOWN" signal is received. The "DOWN" signal is also coupled through switch 1906 to the gate of NMOS transistor 1909. Switch 1906 is used to delay the signal by a time period approximately equal to the delay of inverter 1907. When a positive going signal is received at the "DOWN" terminal, transistor
1909 will turn on. Since transistors 1908 and 1909 are biased by common devices 1912 and 1913, a signal received at the "DOWN" input terminal will steer current between the devices 1908 and 1909. The current through transistor 1909 is also used to drive the output of CP 1900. This current is referred, to as the charge pump sink current.
[0168] Ideally, transistors 1903, 1904, 1910, 1908, 1909, and 1912 should be designed so that the source and sink currents are exactly matched. Thus, if the "UP" and "DOWN" pulses are exactly the same, an equal amount of current will be sourced to and sunk from the output. This ideal case would produce zero static phase offset. However, typically there will always be some device mismatch. Moreover, the bias circuits used to generate PBIAS for transistor
1910 and NBIAS for transistor 1912 may also include some error. An example bias circuit for producing PBIAS and NBIAS is shown in Fig. 21. In this example, the source and sink currents may be adjusted by including transistor 1911 in parallel with transistor 1910, and including transistor 1913 in parallel with transistor 1912. These devices may be used to modify the source and sink currents in transistors 1904 and 1909 to reduce mismatch and the corresponding static phase offset. In this example, an adjust input terminal of CP 1900 is coupled to the gate of PMOS transistor 1914. Transistor 1914 is one-half of a differential pair. The other differential transistor is PMOS device 1915. Transistors 1914 and 1915 are
biased by PMOS transistor 1916, which has a gate coupled to a bias voltage, PBIAS, and a drain coupled to the common sources of transistors 1914 and 1915. The gate of transistor 1915 is coupled to a resistor divider including resistors 1917 and 1918. These resistors set the voltage VD at the gate of device 1915. Current in the drain of transistor 1914 is coupled through current mirrors comprising transistors 1920, 1921, 1922, and 1911 to source coupled transistors 1903 and 1904. Similarly, current in the drain of transistor 1915 is coupled through current mirrors comprising transistors 1930 and 1913 to source coupled transistors 1908 and 1909. Thus, when the adjust input increases above VD, the current in transistor 1913 increases and the current in transistor 1911 decreases, thereby increasing the sink current in transistor 1909 and decreasing the source current in transistor 1904. Similarly, when the adjust input decreases below VD, the current in transistor 1911 increases and the current in transistor 1913 decreases, thereby increasing the source current in transistor 1904 and decreasing the sink current in transistor 1909.
[0169] Consequently, referring to Fig. 18, it can be seen that when a mismatch results in the source current being greater than trie sink current, the voltage on the calibration capacitor will increase and cause comparator 1864A to change states. This will result in a current pulse into capacitor 1875A, which will increase the voltage on the adjust terminal. Increasing the voltage on the adjust terminal will, in turn, decrease the source current and increase the sink current. Accordingly, the mismatch may be reduced. Likewise, it can be seen that when a mismatch results in the sink current being greater than the source current, the voltage on the calibration capacitor will decrease and cause comparator 1865 A to change states. This will result in a current pulse out of capacitor 1875A through transistor 1874A (i.e., a discharge current pulse), which will decrease the voltage on the adjust terminal. Decreasing the voltage on the adjust terminal will, in turn, increase the source current and decrease the sink current. Accordingly, the mismatch error may again be reduced.
[0170] As mentioned above, when phase-frequency detector 1810A and charge pump 181 IA in Fig. 18 are being calibrated, phase-frequency detector 181 OB and charge pump 181 IB are configured to control the timing loop. Additionally, during this time period a second calibration capacitor 1812B is coupled to the output of opamp 1861A and set to a voltage of VCI (the voltage on the loop integration capacitor). Thus, during a second time period phase-frequency detector 1810A and charge pump 181 IA are reconfigured to control the timing loop and phase-frequency detector 181 OB and charge pump 1811 B may be calibrated (i.e., switches 1823, 1824, 1830, 1825, 1826, 1832, 1834 are opened and switches
1821, 1822, 1829, 1831, 1827, 1828, 1833 are closed). In this example, the circuitry for detecting the mismatch in phase-frequency detector 181OB and charge pump 181 IB operates in the same way as the circuitry for detecting the offset in phase-frequency detector 181OA and charge pump 181 IA, wherein the high and low threshold voltages around VCI are again set by the voltages across resistors 1862A-B, and devices 1864- 1875A have corresponding counterparts for 1864-1875B for correcting PFD 1 SlOB and CP 181 IB. Counter 1850 may be used to generate autozero signals, AZ and AZB, to control the switches to alternate between phase detectors. In one embodiment, the modulus for the counter is eight (M=8), so every four cycles the system switches between PFD-CPs. It may be desirable in some implementations to alternate between PFD-CPs after all phase information has been received (e.g., after the phase information from both IREF and IVCO have been received). After both edges have been received, no more timing information is received until the next cycle. Thus, in this example, the counter is driven by both the reference signal IREF and feedback signal IVCO through transmission gates 1853 and 1854, phase-frequency detector 1851, and AND gate 1852. PFD 1851 may be used to ensure that titαe counter only switches the system after both REF and VCO edges have been received. Devices 1853 and 1854 may be used to approximately match the delays of switches 1821, 1822, 1825, and 1826.
[0171] Fig. 22 shows example waveforms for a circuit according to one embodiment of the present invention. Operation of the examples shown in Figs. 17-21 may be understood by referring to the timing diagram shown in Fig. 22. In the timing diagram, the phase-frequency detector and charge pump is assumed to be used in a zero delay buffer application. In the timing diagram of Fig. 22, the PLL is assumed to be very nearly in a locked condition. The inputs have almost no phase difference between them. From the timing diagram, it can be seen that the output of AND gate 1852 (Fig. 18) will be synchronized to VFREF and VCOUT- Thus, CP 181 IA will be in a correction cycle during a time period equal to the first four cycles of VFREF, the CPs will alternate at the end of the fourth cycle of VFREF, and CP 181 IB will be in a correction cycle for a time period equal to the second four cycles of VFREF- It is to be understood that a variety of time periods and techniques for switching between phase detectors could be used. In this example timing diagram, CP 181 IA causes the voltage VCAZI on calibration capacitor 1812A to decrease over time. Accordingly, a pulse VRNI is generated at the gate of NMOS transistor 1874A, which decreases the voltage VCAI on capacitor 1875 A, and therefore decreases the sink current and increases the source current of CP 181 IA. During the next correction cycle for CP 18 HA, it can be seen that the voltage drift of VCAZI
is reduced (i.e., the mismatch error is reduced thereby reducing static phase offset). Similarly, CP 181 IB causes the voltage VCAZ2 on calibration capacitor 1812B to increase over time. Accordingly, a pulse VRP2 is generated at the gate of PMOS transistor 187 IB, which increases the voltage VcA2 on capacitor 1875B, and therefore increases the sink current and decreases the source current of CP 1811 B . During the next correction cycle for CP 181 IB, it can be seen that the voltage drift of VCAZ2 is reduced (i.e., the mismatch error is reduced thereby reducing static phase offset).
[0172] The above description illustrates various embodiments of the present invention along with examples of how aspects of the present invention may be implemented. The above examples and embodiments should not be deemed to be the only embodiments, and are presented to illustrate the flexibility and advantages of the present invention as defined by the following claims. For example, while the above embodiments are described using voltage controlled oscillators or delays, it is to be understood that current controlled devices could also be used. Furthermore, digital implementations of the phase detector, filter, oscillator, or delay could also be used. Based on the above disclosure and the following claims, other arrangements, embodiments, implementations and equivalents will be evident to those skilled in the art and may be employed without departing from the spirit and scope of the invention as defined by the claims.