TWI778886B - Recognition system and sram cell thereof - Google Patents
Recognition system and sram cell thereof Download PDFInfo
- Publication number
- TWI778886B TWI778886B TW110145857A TW110145857A TWI778886B TW I778886 B TWI778886 B TW I778886B TW 110145857 A TW110145857 A TW 110145857A TW 110145857 A TW110145857 A TW 110145857A TW I778886 B TWI778886 B TW I778886B
- Authority
- TW
- Taiwan
- Prior art keywords
- transistor
- capacitor
- output
- inverter
- controlled
- Prior art date
Links
- 239000003990 capacitor Substances 0.000 claims abstract description 59
- 230000003068 static effect Effects 0.000 claims abstract description 11
- 238000005070 sampling Methods 0.000 claims description 42
- 238000013139 quantization Methods 0.000 claims description 14
- 230000000750 progressive effect Effects 0.000 claims description 6
- 238000013528 artificial neural network Methods 0.000 claims description 5
- 230000005540 biological transmission Effects 0.000 claims 1
- 239000013078 crystal Substances 0.000 claims 1
- 238000010586 diagram Methods 0.000 description 49
- 210000004027 cell Anatomy 0.000 description 25
- 230000005669 field effect Effects 0.000 description 12
- 239000004065 semiconductor Substances 0.000 description 11
- 229910044991 metal oxide Inorganic materials 0.000 description 9
- 150000004706 metal oxides Chemical class 0.000 description 9
- 238000012545 processing Methods 0.000 description 6
- 101150103933 VMAC gene Proteins 0.000 description 5
- 101100112673 Rattus norvegicus Ccnd2 gene Proteins 0.000 description 4
- 238000009825 accumulation Methods 0.000 description 4
- 238000001514 detection method Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000003491 array Methods 0.000 description 2
- 241000238558 Eucarida Species 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000000034 method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000012549 training Methods 0.000 description 1
Images
Landscapes
- Investigating Or Analysing Biological Materials (AREA)
- Debugging And Monitoring (AREA)
- Static Random-Access Memory (AREA)
Abstract
Description
本發明係有關識別技術,特別是關於一種使用神經網路的識別系統。The present invention relates to identification technology, in particular to an identification system using a neural network.
語音活動偵測(voice activity detection, VAD)可用以偵測或識別人類的語音。語音活動偵測可觸發基於語音的應用,例如蘋果電腦(Apple)公司的虛擬助理Siri (Speech Interpretation and Recognition Interface,語音解析及辨識介面)。語音活動偵測為一種前端裝置,其通常為一種永開啟(always-on)且為低功率系統。Voice activity detection (VAD) can be used to detect or recognize human speech. Voice activity detection can trigger voice-based applications, such as Apple's virtual assistant Siri (Speech Interpretation and Recognition Interface). Voice activity detection is a front end device, which is usually an always-on and low power system.
現代電腦架構是約翰•馮紐曼(John von Neumann)於1945年提出,但是程式記憶體與資料記憶體之間共享的匯流排(bus)會造成馮紐曼瓶頸。由於單一匯流排僅能於單一時間存取程式記憶體或資料記憶體其中一個,因此處理能力遠低於中央處理器的操作速率。當中央處理器被要求處理大量資料時,將嚴重限制中央處理器的有效處理速率。中央處理器持續被迫等待,才能將所需資料移至記憶體或從記憶體取出。Modern computer architecture was proposed by John von Neumann in 1945, but the shared bus between program memory and data memory creates a von Neumann bottleneck. Since a single bus can only access either the program memory or the data memory at a single time, the processing capability is much lower than the operating speed of the central processing unit. When the central processing unit is required to process a large amount of data, the effective processing rate of the central processing unit will be severely limited. The central processing unit is constantly forced to wait before it can move the required data to or from memory.
記憶體內運算(Computing-in-Memory, CIM)係一種整合運算與記憶體的技術。將運算嵌入記憶體可減少資料的搬移,使得能量的使用更有效率,且大量並行操作可節省頻寬。記憶體內運算有助於邊緣運算(edge computing),其為一種將運算與資料儲存靠近資料源的分散式運算方式,可應用於物聯網(Internet of Things, IoT)的機器學習。Computing-in-Memory (CIM) is a technology that integrates computing and memory. Embedding operations into memory reduces data movement, making energy use more efficient, and saving bandwidth for massively parallel operations. In-memory computing contributes to edge computing, which is a decentralized computing method that stores computing and data close to the data source, and can be applied to machine learning in the Internet of Things (IoT).
因此,亟需提出一種新穎機制,以增進低功率或/且高頻寬系統(例如語音識別系統)的效能。Therefore, there is a need for a novel mechanism to improve the performance of low power or/and high bandwidth systems such as speech recognition systems.
鑑於上述,本發明實施例的目的之一在於提出一種包含靜態隨機存取記憶體(Static Random Access Memory, SRAM)單元(cell)的識別系統,其採用電荷重新分配(charge redistribution)原理以產生累加信號,具低功耗或/且提升頻寬。In view of the above, one of the objectives of the embodiments of the present invention is to provide an identification system including a Static Random Access Memory (SRAM) cell, which adopts the principle of charge redistribution to generate accumulation signal with low power consumption or/and increased bandwidth.
根據本發明實施例,識別系統包含複數靜態隨機存取記憶體單元及量化器。靜態隨機存取記憶體單元排列成行,每一行的靜態隨機存取記憶體單元分別接收相應輸入信號並產生相應輸出信號,其相連接以產生子信號,且所有行的子信號相連接以產生累加信號。量化器接收累加信號以產生數位輸出,該量化器包含至少一電容器陣列,其與靜態隨機存取記憶體單元共享。According to an embodiment of the present invention, the identification system includes a complex SRAM unit and a quantizer. The SRAM cells are arranged in rows, the SRAM cells of each row respectively receive corresponding input signals and generate corresponding output signals, which are connected to generate sub-signals, and the sub-signals of all rows are connected to generate accumulation Signal. A quantizer receives the accumulated signal to generate a digital output, the quantizer includes at least one capacitor array, which is shared with the SRAM cell.
第一A圖顯示本發明實施例的適用於識別系統的(人工)神經網路的示意圖。神經網路可包含相連的節點(或神經元),相連節點之間具權重,其可藉由資料集(dataset)經訓練獲得。識別系統可適用於語音識別,以識別輸入信號究竟為語音或雜訊。如第一A圖所例示,輸入層的節點接收輸入信號(例如Vin1~Vin6),分別代表不同通道的萃取特徵,且輸出層的節點可識別輸入信號究竟為語音或雜訊。一或多隱藏層(例如所示的第一層)的節點接收前一層的輸出,據以產生輸出以發送至後一層。The first figure A shows a schematic diagram of an (artificial) neural network suitable for a recognition system according to an embodiment of the present invention. A neural network can include connected nodes (or neurons) with weights between the connected nodes, which can be obtained by training a dataset. The recognition system can be adapted for speech recognition to recognize whether the input signal is speech or noise. As illustrated in Figure A, the nodes of the input layer receive input signals (eg Vin1~Vin6), which represent the extraction features of different channels, and the nodes of the output layer can identify whether the input signal is speech or noise. Nodes of one or more hidden layers (eg, the first layer shown) receive the output of the previous layer, and thereby generate output to send to the subsequent layer.
第一B圖顯示本發明實施例的識別系統100的方塊圖。第一B圖所例示的識別系統100可適用於第一A圖的輸入層。根據本實施例的特徵之一,識別系統100使用記憶體內運算(CIM)技術,其整合運算與記憶體,因而降低功耗且節省頻寬。The first diagram B shows a block diagram of the
本實施例的識別系統100可包含複數靜態隨機存取記憶體(SRAM)單元11,排列成行。對於每一行,靜態隨機存取記憶體單元11分別接收相應輸入信號(例如Vin1~Vin6)並產生相應(加權)輸出信號,其相連接(相加)以產生子信號(sub-signal)。接著,所有行的子信號相連接以產生累加信號Vmac,其代表(該複數靜態隨機存取記憶體單元11)對輸入信號進行乘積累加(multiply-accumulate)運算所得到的輸出信號。The
第二A圖顯示本發明實施例的靜態隨機存取記憶體單元11(第一B圖)的電路圖。在本實施例中,靜態隨機存取記憶體單元11可包含八個電晶體(例如金屬氧化物半導體場效電晶體(MOSFET))與一個電容器(亦即,8T1C)。本實施例的靜態隨機存取記憶體單元11可包含第一反向器,其包含第一電晶體M1(例如N型金屬氧化物半導體場效電晶體)與第二電晶體M2(例如P型金屬氧化物半導體場效電晶體),串聯於地與電源之間,其中第二電晶體M2的型態相反於第一電晶體M1。靜態隨機存取記憶體單元11還包含第二反向器,其包含第三電晶體M3(例如N型金屬氧化物半導體場效電晶體)與第四電晶體M4(例如P型金屬氧化物半導體場效電晶體),串聯於地與電源之間,其中第四電晶體M4的型態相反於第三電晶體M3。第一反向器(M1、M2)與第二反向器(M3、M4)交叉耦接(cross-coupled)。亦即,第一反向器(M1、M2)的輸出Q耦接至第二反向器(M3、M4)的輸入,且第二反向器(M3、M4)的(反向)輸出Qb耦接至第一反向器(M1、M2)的輸入。The second diagram A shows a circuit diagram of the SRAM cell 11 (the first diagram B) according to the embodiment of the present invention. In this embodiment, the
靜態隨機存取記憶體單元11可包含第一存取電晶體,其包含第五電晶體M5(例如N型金屬氧化物半導體場效電晶體),受控於字元線(word line)WL以存取第一反向器(M1、M2)的輸出Q,其藉由第一位元線(bit line)BL以進行傳送。靜態隨機存取記憶體單元11還包含第二存取電晶體,其包含第六電晶體M6(例如N型金屬氧化物半導體場效電晶體),受控於字元線WL以存取第二反向器(M3、M4)的(反向)輸出Qb,其藉由第二位元線BLb以進行傳送。The
根據本實施例的特徵之一,靜態隨機存取記憶體單元11可包含第一通路電晶體,其包含第七電晶體M7(例如N型金屬氧化物半導體場效電晶體),(其閘極)受控於第一反向器(M1、M2)的輸出Q,用以讓共模(common-mode)電壓Vcm(經由汲極)通過。靜態隨機存取記憶體單元11還包含第二通路電晶體,其包含第八電晶體M8(例如N型金屬氧化物半導體場效電晶體),(其閘極)受控於第二反向器(M3、M4)的輸出Qb,用以讓輸入信號Vin(經由汲極)通過。第七電晶體M7與第八電晶體M8的輸出(於源極)連接在一起。According to one of the features of the present embodiment, the
根據本實施例的另一特徵,靜態隨機存取記憶體單元11可包含電容器C,分別藉由第七電晶體M7與第八電晶體M8,切換(switchably)耦接共模電壓Vcm與輸入信號Vin。在本實施例中,電容器C藉由開關SW1以切換接收第一/第二通路電晶體M7/M8的輸出,該開關SW1受控於取樣時脈信號CLKs。值得注意的是,(識別系統100的)不同行的電容器C的值(或權重)彼此不同。其中,不同行的靜態隨機存取記憶體單元11的電容器C分別具有加權二進位(binary-weighted)值(例如C、2C、4C及8C)。According to another feature of the present embodiment, the
第二B圖顯示本發明另一實施例的靜態隨機存取記憶體單元11(第一B圖)的電路圖。在本實施例中,靜態隨機存取記憶體單元11可包含十個電晶體與一個電容器(亦即,10T1C)。第二B圖的靜態隨機存取記憶體單元11類似於第二A圖,相異處說明如下。如第二B圖所示,靜態隨機存取記憶體單元11更包含第一切換電晶體,其包含第九電晶體M9(例如N型金屬氧化物半導體場效電晶體),串聯於第一通路電晶體M7;及第二切換電晶體,其包含第十電晶體M10(例如N型金屬氧化物半導體場效電晶體),串聯於第二通路電晶體M8。藉此,第一通路電晶體M7藉由第一切換電晶體M9(其閘極受控於取樣時脈信號CLKs),以間接接收共模電壓Vcm;且第二通路電晶體M8藉由第二切換電晶體M10(其閘極受控於取樣時脈信號CLKs),以間接接收輸入信號Vin。然而,電容器C不藉由開關SW1而直接連接至第一/第二通路電晶體M7/M8的輸出。因此,第一切換電晶體M9與第二切換電晶體M10共同作為一個開關,藉以讓電容器C可以分別藉由第一通路電晶體M7與第二通路電晶體M8,切換接收共模電壓Vcm與輸入信號Vin。The second diagram B shows a circuit diagram of the SRAM cell 11 (the first diagram B) according to another embodiment of the present invention. In this embodiment, the
於操作時,開關SW1(或第一/第二切換電晶體M9/M10)於取樣階段為導通,使得電容器C的下板(從第一/第二通路電晶體M7/M8的輸出)取樣類比電壓,而電容器C的上板則耦接至共模電壓Vcm。開關SW1(或第一/第二切換電晶體M9/M10)於量化(quantization)階段為斷開,電容器C的下板藉由反向開關SW2,切換耦接至參考電壓Vref(例如正參考電壓Vrefp或負參考電壓Vrefn),該反向開關SW2受控於反向取樣時脈信號CLKsb(其相對於取樣時脈信號CLKs具有相反極性),而電容器C的上板則得到先前的取樣電壓。In operation, the switch SW1 (or the first/second switching transistor M9/M10) is turned on during the sampling phase, so that the lower plate of the capacitor C (from the output of the first/second pass transistor M7/M8) is sampled analogously voltage, and the upper plate of the capacitor C is coupled to the common mode voltage Vcm. The switch SW1 (or the first/second switching transistor M9/M10) is turned off in the quantization stage, and the lower plate of the capacitor C is switched to be coupled to the reference voltage Vref (eg, a positive reference voltage) through the reverse switch SW2 Vrefp or negative reference voltage Vrefn), the reverse switch SW2 is controlled by the reverse sampling clock signal CLKsb (which has the opposite polarity with respect to the sampling clock signal CLKs), and the upper plate of the capacitor C obtains the previous sampling voltage.
第三A圖顯示本發明實施例的循續漸近式類比至數位轉換器(successive approximation register analog-to-digital converter, SAR ADC)200A的電路圖,作為識別系統100當中的量化器。循續漸近式類比至數位轉換器200A將連續的類比波轉換為分離的數位值,其於所有量化位準執行二元搜尋(binary search),於每一次轉換最終收斂得到數位輸出Dout。循續漸近式類比至數位轉換器200A係作為量化器,於操作時可配合第一B圖的靜態隨機存取記憶體單元11。循續漸近式類比至數位轉換器200A可包含第一數位至類比轉換器(DAC)21,包含電容器陣列;與第二數位至類比轉換器(DAC)22,包含電容器陣列。第一數位至類比轉換器(DAC)21與第二數位至類比轉換器(DAC)22的電容器陣列可切換耦接至輸入信號(例如Vin1~Vin6)與共模電壓Vcm。循續漸近式類比至數位轉換器200A可包含比較器23,其(於非反向輸入節點)接收第一數位至類比轉換器(DAC)21的輸出,且(於反向輸入節點)接收第二數位至類比轉換器(DAC)22的輸出。此外,比較器23還接收累加信號Vmac。循續漸近式類比至數位轉換器200A可包含循續漸近式邏輯24,接收比較器23的比較結果,據以產生數位輸出Dout。根據本實施例的特徵之一,第一數位至類比轉換器(DAC)21與第二數位至類比轉換器(DAC)22的電容器陣列可與第一B圖的靜態隨機存取記憶體單元11共享。FIG. 3 A shows a circuit diagram of a successive approximation register analog-to-digital converter (SAR ADC) 200A as a quantizer in the
第三B圖顯示本發明另一實施例的循續漸近式類比至數位轉換器(SAR ADC)200B的電路圖,作為識別系統100當中的量化器。第三B圖的循續漸近式類比至數位轉換器(SAR ADC)200B類似於第三A圖的循續漸近式類比至數位轉換器(SAR ADC)200A,其差異處說明如下。在本實施例中,第一數位至類比轉換器(DAC)21與第二數位至類比轉換器(DAC)22更分別包含第一虛擬(dummy或複製)電容器211與第二虛擬(dummy或複製)電容器221,其切換連接至輸入信號(例如Vin1~Vin6)與共模電壓Vcm。FIG. 3 B shows a circuit diagram of a successive asymptotic analog-to-digital converter (SAR ADC) 200B as a quantizer in the
第四A圖顯示循續漸近式類比至數位轉換器(SAR ADC)200B於第一取樣階段的等效電路,且第四B圖顯示循續漸近式類比至數位轉換器(SAR ADC)200B於第二取樣階段的等效電路。第四C圖與第四D圖分別顯示循續漸近式類比至數位轉換器(SAR ADC)200B於第一量化階段與第二量化階段且當(第一數位至類比轉換器(DAC)21的)Vip大於(第二數位至類比轉換器(DAC)22的)Vin時的等效電路。第四E圖與第四F圖分別顯示循續漸近式類比至數位轉換器(SAR ADC)200B於第一量化階段與第二量化階段且當(第一數位至類比轉換器(DAC)21的)Vip小於(第二數位至類比轉換器(DAC)22的)Vin時的等效電路。Figure 4 A shows the equivalent circuit of the
根據上述實施例,靜態隨機存取記憶體單元11採用電荷重新分配(charge redistribution)原理以產生累加信號Vmac,而非使用傳統系統的電荷分享(charge sharing)原理。因此,相較於傳統系統,本發明上述實施例的時序變得簡單,且不需要重置階段。此外,循續漸近式類比至數位轉換器200A/B的電容器陣列可於取樣階段以產生累加信號Vmac。再者,於第三B圖的循續漸近式類比至數位轉換器200B使用虛擬電容器,使得信號擺幅可接近全部範圍。According to the above-mentioned embodiment, the
第五A圖顯示數位至類比轉換器(DAC)500的電路圖,其代表第一A圖的第一層的節點。第五B圖與第五C圖分別顯示數位至類比轉換器500於重置階段與輸出階段的等效電路圖,且第五D圖例示第五B圖與第五C圖的相關信號的時序圖。在本實施例中,數位至類比轉換器(DAC)500可包含電容器陣列,其包含複數電容器(例如C、2C、4C及8C)。電容器的上板連接在一起作為數位至類比轉換器(DAC)500的輸出DACout。電容器的下板切換接收前一層的數位輸出或反向數位輸出。其中,bit0~bit3代表前一層(例如第一A圖的輸入層)的循續漸近式邏輯24所產生的數位輸出,且bit0b~bit3b代表反向數位輸出,其極性分別相反於數位輸出bit0~bit3。舉例而言,數位輸出bit0與bit0b分別藉由開關SW與SWb而電性耦接至相應電容器(其中開關SW的操作相反於開關SWb),該開關SW/SWb受控於事先訓練與事先儲存的權重。接著,數位至類比轉換器(DAC)500的輸出DACout傳送至比較器(例如位於輸出層的節點),據以識別語音或雜訊。The fifth A-diagram shows a circuit diagram of a digital-to-analog converter (DAC) 500, which represents the nodes of the first layer of the first A-diagram. The fifth diagram B and the fifth diagram C show the equivalent circuit diagrams of the digital-to-
以上所述僅為本發明之較佳實施例而已,並非用以限定本發明之申請專利範圍;凡其它未脫離發明所揭示之精神下所完成之等效改變或修飾,均應包含在下述之申請專利範圍內。The above descriptions are only preferred embodiments of the present invention and are not intended to limit the scope of the patent application of the present invention; all other equivalent changes or modifications made without departing from the spirit disclosed in the invention shall be included in the following within the scope of the patent application.
100:識別系統
11:靜態隨機存取記憶體單元
200A:循續漸近式類比至數位轉換器
200B:循續漸近式類比至數位轉換器
21:第一數位至類比轉換器
211:第一虛擬電容器
22:第二數位至類比轉換器
221:第二虛擬電容器
23:比較器
24:循續漸近式邏輯
Vin:輸入信號
Vin1~Vin6:輸入信號
CLKs:取樣時脈信號
CLKsb:反向取樣時脈信號
Vcm:共模電壓
Vmac:累加信號
SRAM:靜態隨機存取記憶體
M1:第一電晶體
M2:第二電晶體
M3:第三電晶體
M4:第四電晶體
M5:第五電晶體/第一存取電晶體
M6:第六電晶體/第二存取電晶體
M7:第七電晶體/第一通路電晶體
M8:第八電晶體/第二通路電晶體
M9:第九電晶體/第一切換電晶體
M10:第十電晶體/第二切換電晶體
Q:輸出
Qb:輸出
BL:第一位元線
BLb:第二位元線
WL:字元線
SW1:開關
SW2:反向開關
C:電容器
Vref:參考電壓
Vrefp:正參考電壓
Vrefn:負參考電壓
Dout:數位輸出
DACout:輸出
bit0~bit3:數位輸出
bit0b~bit3b:反向數位輸出
SW:開關
SWb:開關
Reset:重置信號
Reset_b:反向重置信號
100: Identification System
11: Static random
第一A圖顯示本發明實施例的適用於識別系統的(人工)神經網路的示意圖。 第一B圖顯示本發明實施例的識別系統的方塊圖。 第二A圖顯示本發明實施例的靜態隨機存取記憶體單元 (第一B圖)的電路圖。 第二B圖顯示本發明另一實施例的靜態隨機存取記憶體單元(第一B圖)的電路圖。 第三A圖顯示本發明實施例的循續漸近式類比至數位轉換器(SAR ADC)的電路圖,作為識別系統當中的量化器。 第三B圖顯示本發明另一實施例的循續漸近式類比至數位轉換器(SAR ADC)的電路圖,作為識別系統當中的量化器。 第四A圖顯示循續漸近式類比至數位轉換器(SAR ADC)於第一取樣階段的等效電路。 第四B圖顯示循續漸近式類比至數位轉換器(SAR ADC)於第二取樣階段的等效電路。 第四C圖與第四D圖分別顯示循續漸近式類比至數位轉換器(SAR ADC)於第一量化階段與第二量化階段且當(第一數位至類比轉換器(DAC)的)Vip大於(第二數位至類比轉換器(DAC)的)Vin時的等效電路。 第四E圖與第四F圖分別顯示循續漸近式類比至數位轉換器(SAR ADC)於第一量化階段與第二量化階段且當(第一數位至類比轉換器(DAC)的)Vip小於(第二數位至類比轉換器(DAC)的)Vin時的等效電路。 第五A圖顯示數位至類比轉換器(DAC)的電路圖,其代表第一A圖的第一層的節點。 第五B圖與第五C圖分別顯示數位至類比轉換器於重置階段與輸出階段的等效電路圖。 第五D圖例示第五B圖與第五C圖的相關信號的時序圖。 The first figure A shows a schematic diagram of an (artificial) neural network suitable for a recognition system according to an embodiment of the present invention. The first Figure B shows a block diagram of an identification system according to an embodiment of the present invention. The second diagram A shows a circuit diagram of the SRAM cell (the first diagram B) according to an embodiment of the present invention. The second diagram B shows a circuit diagram of a SRAM cell (the first diagram B) according to another embodiment of the present invention. Figure 3 A shows a circuit diagram of a successive asymptotic analog-to-digital converter (SAR ADC) according to an embodiment of the present invention as a quantizer in a recognition system. Figure 3 B shows a circuit diagram of a successive asymptotic analog-to-digital converter (SAR ADC) as a quantizer in an identification system according to another embodiment of the present invention. Figure 4A shows the equivalent circuit of the successive asymptotic analog-to-digital converter (SAR ADC) in the first sampling stage. Figure 4 B shows the equivalent circuit of the successive asymptotic analog-to-digital converter (SAR ADC) in the second sampling stage. The fourth graph C and the fourth graph D show the successive asymptotic analog-to-digital converter (SAR ADC) in the first quantization stage and the second quantization stage respectively and when the Vip (of the first digital-to-analog converter (DAC)) Equivalent circuit when greater than Vin (of the second digital-to-analog converter (DAC)). The fourth diagram E and the fourth diagram F show the successive asymptotic analog-to-digital converter (SAR ADC) in the first quantization stage and the second quantization stage respectively and when the Vip (of the first digital-to-analog converter (DAC)) Equivalent circuit for less than Vin (of the second digital-to-analog converter (DAC)). The fifth A-diagram shows a circuit diagram of a digital-to-analog converter (DAC) representing the nodes of the first layer of the first A-diagram. The fifth diagram B and the fifth diagram C show the equivalent circuit diagrams of the digital-to-analog converter in the reset stage and the output stage, respectively. The fifth D diagram illustrates the timing diagram of the correlation signals of the fifth B diagram and the fifth C diagram.
11:靜態隨機存取記憶體單元 11: Static random access memory unit
Vin:輸入信號 Vin: input signal
CLKs:取樣時脈信號 CLKs: sampling clock signal
CLKsb:反向取樣時脈信號 CLKsb: reverse sampling clock signal
Vcm:共模電壓 Vcm: common mode voltage
M1:第一電晶體 M1: first transistor
M2:第二電晶體 M2: second transistor
M3:第三電晶體 M3: The third transistor
M4:第四電晶體 M4: Fourth transistor
M5:第五電晶體/第一存取電晶體 M5: Fifth transistor/first access transistor
M6:第六電晶體/第二存取電晶體 M6: sixth transistor/second access transistor
M7:第七電晶體/第一通路電晶體 M7: seventh transistor/first pass transistor
M8:第八電晶體/第二通路電晶體 M8: Eighth Transistor/Second Pass Transistor
Q:輸出 Q: output
Qb:輸出 Qb: output
BL:第一位元線 BL: first element line
BLb:第二位元線 BLb: Second bit line
WL:字元線 WL: word line
SW1:開關 SW1: switch
SW2:反向開關 SW2: reverse switch
C:電容器 C: capacitor
Vref:參考電壓 Vref: reference voltage
Claims (16)
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
TW110145857A TWI778886B (en) | 2021-12-08 | 2021-12-08 | Recognition system and sram cell thereof |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
TW110145857A TWI778886B (en) | 2021-12-08 | 2021-12-08 | Recognition system and sram cell thereof |
Publications (2)
Publication Number | Publication Date |
---|---|
TWI778886B true TWI778886B (en) | 2022-09-21 |
TW202324407A TW202324407A (en) | 2023-06-16 |
Family
ID=84958387
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
TW110145857A TWI778886B (en) | 2021-12-08 | 2021-12-08 | Recognition system and sram cell thereof |
Country Status (1)
Country | Link |
---|---|
TW (1) | TWI778886B (en) |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080224196A1 (en) * | 2007-03-15 | 2008-09-18 | Nec Electronics Corporation | Semiconductor device and manufacturing process for the same |
US20150103585A1 (en) * | 2013-10-11 | 2015-04-16 | United Microelectronics Corp. | High stability static random access memory cell |
US10276578B2 (en) * | 2017-06-25 | 2019-04-30 | United Microelectronics Corp. | Dynamic oxide semiconductor random access memory(DOSRAM) having a capacitor electrically connected to the random access memory (SRAM) |
US20210343343A1 (en) * | 2020-04-30 | 2021-11-04 | Qualcomm Incorporated | Voltage offset for compute-in-memory architecture |
US20210366543A1 (en) * | 2020-02-10 | 2021-11-25 | Taiwan Semiconductor Manufacturing Company, Ltd. | Memory with fram and sram of ic |
-
2021
- 2021-12-08 TW TW110145857A patent/TWI778886B/en active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080224196A1 (en) * | 2007-03-15 | 2008-09-18 | Nec Electronics Corporation | Semiconductor device and manufacturing process for the same |
US20150103585A1 (en) * | 2013-10-11 | 2015-04-16 | United Microelectronics Corp. | High stability static random access memory cell |
US10276578B2 (en) * | 2017-06-25 | 2019-04-30 | United Microelectronics Corp. | Dynamic oxide semiconductor random access memory(DOSRAM) having a capacitor electrically connected to the random access memory (SRAM) |
US20210366543A1 (en) * | 2020-02-10 | 2021-11-25 | Taiwan Semiconductor Manufacturing Company, Ltd. | Memory with fram and sram of ic |
US20210343343A1 (en) * | 2020-04-30 | 2021-11-04 | Qualcomm Incorporated | Voltage offset for compute-in-memory architecture |
Also Published As
Publication number | Publication date |
---|---|
TW202324407A (en) | 2023-06-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Biswas et al. | Conv-RAM: An energy-efficient SRAM with embedded convolution computation for low-power CNN-based machine learning applications | |
EP3989445A1 (en) | Sub-unit, mac array, bit-width reconfigurable hybrid analog-digital in-memory computing module | |
CN111967586B (en) | Chip for pulse neural network memory calculation and calculation method | |
CN110378475B (en) | A neuromorphic computing circuit based on a multi-bit parallel binary synapse array | |
CN104967451B (en) | Gradual approaching A/D converter | |
US20220301605A1 (en) | Compute-in-memory with ternary activation | |
CN106067817A (en) | 1.5 redundancy bits based on controlled asymmetric dynamic comparator accelerate gradual approaching A/D converter | |
CN111865319A (en) | An ultra-low power successive approximation analog-to-digital converter based on a four-input comparator | |
JPH0734541B2 (en) | Successive approximation type analog-digital conversion system | |
TWI752823B (en) | Memory system | |
TWI803889B (en) | Computing device and computing method | |
US20230161627A1 (en) | High-energy-efficiency binary neural network accelerator applicable to artificial intelligence internet of things | |
CN115910152A (en) | Charge domain memory calculation circuit and calculation circuit with positive and negative number operation function | |
Cao et al. | Evaluating neural network-inspired analog-to-digital conversion with low-precision RRAM | |
TWI778886B (en) | Recognition system and sram cell thereof | |
Xuan et al. | High-efficiency data conversion interface for reconfigurable function-in-memory computing | |
CN119171907A (en) | A multi-step analog-to-digital converter with a common reference voltage and an operation method thereof | |
CN117063151A (en) | Charge domain in-memory computing circuit | |
CN116264076A (en) | Identify the system and its static random access memory cells | |
US11990178B2 (en) | Recognition system and SRAM cell thereof | |
Zhang et al. | A 28nm 15.09 nJ/inference Neuromorphic Processor with SRAM-Based Charge Domain in-Memory-Computing | |
CN113162624B (en) | Analog-to-digital converter with pooling function | |
CN119271172B (en) | Charge domain signed multiplication, multi-bit multiplication and accumulation circuit and chip thereof | |
Liu et al. | Design of switched-current based low-power PIM vision system for IoT applications | |
CN119692261B (en) | Hybrid ADC circuit and module for charge domain SRAM memory computation |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
GD4A | Issue of patent certificate for granted invention patent |