A Successive Approximation Analog-to-Digital Converter Circuit for Analog In-Memory Computation
By designing a successive approximation analog-to-digital converter circuit, using charge redistribution and output offset storage technology, the problems of high power consumption and large area in analog in-memory computing are solved, and the power consumption and area under high precision requirements are achieved Optimization improves computing efficiency.
Patent Information
- Application Number
- CN202210104636.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-01-28
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2042-01-28
AI Technical Summary
In existing analog in-memory computing, analog-to-digital converters (especially ADCs) have high power consumption and large area occupancy, resulting in poor computing efficiency and density advantages in data scenarios with low accuracy requirements. In addition, traditional flash ADCs have Power consumption and area costs increase significantly when improving accuracy.
A successive approximation analog-to-digital converter circuit is designed, including an 8-bit charge redistribution digital-to-analog converter circuit, a current bias circuit with a start circuit, a voltage-to-analog reference circuit controlled by the op amp and a high-speed latch comparator circuit. , combined with the binary search algorithm to achieve the analog-to-digital conversion function of successive approximation, and through the capacitor array and complementary CMOS switch, output offset storage technology is used to reduce static power consumption.
Under the high precision requirements of 8-bit codewords, compared with traditional flash analog-to-digital converters, it has the advantages of power consumption and area, effectively controls the leakage of the current bias module, and improves the linearity and low analog-to-digital converters. Power consumption characteristics.
Smart Images

Figure CN114759927B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of integrated circuit design, and in particular to a successive approximation analog-to-digital converter circuit that can reduce power consumption and save area overhead in various analog in-memory computing application scenarios. Background Art
[0002] Traditional computing systems based on the von Neumann architecture are fundamentally bottlenecked by data transfer between the processor and memory. As processor performance increases rapidly and far outpaces memory speed, the memory access bottleneck is becoming increasingly prominent. The emergence of data-intensive tasks such as machine learning urgently requires a solution to this bottleneck by designing a solution that allows processors and memory to perform computations on the same platform. This solution, called "in-memory computing," can potentially eliminate the cost of data access and transmission by performing computations directly inside the memory array. Crossbar arrays based on resistive non-volatile memory (NVM) show great promise in in-memory computing systems used for machine learning workloads because their high density can lead to higher on-chip storage capacity, while they can also perform massively parallel matrix-vector multiplication (MVM) operations, thereby accelerating the main computing core of machine learning workloads. Due to device-level and circuit-level non-idealities, analog computing based on resistive crossbar arrays is essentially approximate. In addition, the area and power costs of peripheral circuits used to convert between analog and digital domains greatly reduce the intrinsic efficiency of MVM computing based on NVM crossbar arrays.
[0003] The output calculated by the analog MVM unit needs to be transmitted to other computing units to form a large-scale neural network. Since analog signals have a low tolerance to noise, signal transmission in any large network needs to be digitized. Therefore, the MVM computing unit requires peripheral circuits, such as digital-to-analog converters (DACs) at the input and analog-to-digital converters (ADCs) at the output. These peripheral circuits in the MVM unit, especially the ADC, account for a large part of the total energy consumption of the MVM unit in most existing works. In some works, the power consumption of the ADC is as high as 80% of the MVM unit and the area occupied is close to 70% of the MVM unit. Therefore, the power consumption and area of the ADC as a peripheral circuit significantly reduce the advantages of the crossbar array in terms of computational efficiency and high density. Since the power consumption of the ADC is positively correlated with its accuracy in function, we can take advantage of the tolerance of the neural network to errors and strike a balance between the accuracy and power consumption of the NVM crossbar array by properly configuring the accuracy of the ADC. When facing some data scenarios with low accuracy requirements, we should try to pursue low power consumption and area-optimized ADC peripheral circuits.
[0004] Attached Figure 1The following is a structural block diagram of a common analog in-memory computing mode in the prior art. The memory cells that make up the MVM crossbar array can be classic static random access memory (SRAM), dynamic random access memory (DRAM) or flash memory (Flash) array composed of floating gate transistors, or some new non-volatile memory (NVM), such as memristor (ReRAM), ferroelectric memory (FeRAM), phase change random access memory (PCM) or magnetic random access memory (MRAM). As shown in the attached figure, Figure 1 As shown in the figure, the structure in the dashed box in the upper left corner is a traditional memory structure, which includes a cell array, a read / write circuit, and a drive circuit. The mode controller can switch this type of integrated memory and computing chip from storage mode to in-memory computing mode. Figure 1 The module in the dotted box in the lower right corner starts working. Taking the MVM acceleration of convolutional neural network as an example, the weights are generally recorded in the unit array in the form of resistance values according to the different storage media, and the digital activation X IN The input is converted into an analog signal through the DAC and enters the array. In the array, the activation and weight are convolved in the form of analog electrical signals, and then converted into digital signals through the multiplexer (MUX) and ADC, and the formal digital result Y is obtained by shift addition. OUT Then enter the next calculation unit.
[0005] Attached Figure 2 The schematic diagram of the structure of a matrix-vector multiplier circuit based on a cross array of flash memory devices in the prior art. In the working mode of this NOR-type Flash storage and computing integrated chip, the analog signal is input into the array from the bit line (BL), and the weight value of the neural network is based on the gate threshold voltage V TH The floating gate transistor is driven by the word line (WL), and the output current of the floating gate transistor is equal to the product of the input current and the weight. Since the output is current rather than voltage, the accumulation operation can be easily performed without using additional logic gates. In other words, the calculation result is naturally output on the source line (SL) under the action of Kirchhoff's current law.
[0006] Attached Figure 3 The figure is a schematic diagram of the structure of a conventional flash analog-to-digital converter (Flash ADC) circuit in the prior art. Flash ADC is the simplest type of conversion method. Since each bit of its codeword is converted simultaneously during operation, it is also called a fully parallel ADC. Therefore, it is also the ADC with the fastest conversion speed among all types. The working principle of N-bit flash ADC is to convert a total of 2 N The resistor network is composed of a series voltage divider. The resistor network divides the reference voltage V REF Divide into 2 N voltage, and then these 2 N Voltage input to a total of 2N -1 comparator array as a reference voltage, the reference voltage and the input voltage V IN Comparison, a string of temperature codes obtained is digitally output through decoders and latches. Although the flash ADC has a very fast conversion speed, since the conversion is fully parallel, it requires high matching accuracy between comparators, and mismatch will cause large errors. On the other hand, the number of resistors and comparators required by the flash ADC increases exponentially with the number of bits of resolution. When the accuracy is increased above 4 bits, it means a huge increase in area and power consumption, which is obviously not applicable in the pursuit of low power consumption and high precision in-memory calculations. Summary of the invention
[0007] Purpose of the invention: In view of the above-mentioned deficiencies in the prior art, the present invention provides a successive approximation analog-to-digital converter circuit for analog in-memory calculations, especially for use in integrated memory and computing chips that have certain requirements on ADC accuracy, power consumption and area.
[0008] Technical solution: The purpose of the present invention is achieved through the following technical solutions:
[0009] In order to achieve the above-mentioned invention object, the present invention provides a successive approximation analog-to-digital converter circuit for analog in-memory calculation, which includes the following modules: an 8-bit charge redistribution digital-to-analog converter circuit, a current bias circuit with a startup circuit, a voltage analog reference circuit controlled by an operational amplifier, a high-speed latch comparator circuit, and a successive approximation logic control module for implementing a binary search algorithm. The current bias circuit is connected to a reference circuit, a buffer circuit attached to the reference circuit, and a comparator circuit to provide a current or voltage bias; the digital-to-analog converter circuit is controlled by a digital logic control module for digital input, and a high-speed latch comparator is used to compare the analog input with the reference voltage, and the comparison result is fed back to the digital input of the digital-to-analog converter for adjustment, and finally the analog-to-digital conversion function of successive approximation is realized in the form of a binary search algorithm.
[0010] Further, in the present invention: the 8-bit charge redistribution type digital-to-analog converter circuit is composed of a capacitor array C0-C8, a bridge capacitor Cc, and complementary CMOS switches S0-S8. The capacitors C1-C8 correspond to the lowest bit to the highest bit codeword of the 8-bit digital-to-analog conversion respectively, and the capacitor C0 is a complementary capacitor, and the capacitance is the same as the capacitor C1 of the least significant bit (LSB). In order to reduce the area of the capacitor array, the bridge capacitor Cc is connected in series between the high 4-bit capacitor and the low 4-bit capacitor, and the capacitance ratio of the capacitors C0-C8 and Cc is 1:1:2:4:8:2:4:8:16:16 / 7. The lower plates of the capacitor array C0-C8 are connected to the switches S0-S8 respectively, and the upper plates of the capacitor array, as the output end of the entire 8-bit charge redistribution type digital-to-analog converter circuit, are connected to the positive input end of the latch comparator. Switches S0~S8 are controlled by the output signals Sample and B[7:0] of the successive approximation logic control circuit. The control signal Sample simultaneously regulates the switching between the analog input and the reference voltage of the entire analog-to-digital converter, as well as the resetting of the common-mode level in the upper plate of the capacitor array and the latch comparator.
[0011] Furthermore, in the present invention: the current bias circuit with a startup circuit is composed of two pairs of current mirrors forming a loop and a resistor to form a classic monotonic △V GS / R type self-bias circuit, wherein the PMOS tube current mirror and the NMOS tube current mirror provide current bias for the reference circuit in the analog-to-digital converter and the buffer circuit after the reference circuit respectively, and then use an NMOS tube and a PMOS tube to copy the current to the gate voltage on the PMOS tube, provide voltage bias for the high-speed comparator in the analog-to-digital converter and isolate the influence of the subsequent stage. On this basis, the self-bias circuit has a set of robust startup circuits and has low leakage when the enable is zero.
[0012] Furthermore, in the present invention: the voltage-mode reference circuit controlled by the operational amplifier, based on the principle of the classic reference voltage source, uses two pairs of PMOS tubes and a resistor to form a common source and common gate structure, and adds another pair of PMOS tubes to enable the reference source. The two-stage operational amplifier used in the reference source also adds an NMOS tube and a PMOS tube to enable the two stages of the operational amplifier respectively.
[0013] Furthermore, in the present invention: the high-speed latch comparator circuit is composed of three latch comparison stages and one output buffer stage. The three latch comparison stages are static latch circuits with the same structure, which are used to compare and determine the analog voltage at the positive end of the input comparator with the common mode level at the negative end, and then output the logic level. The output buffer stage circuit is a simple differential operational amplifier, which provides sufficient gain for the comparator and outputs a stable logic level. The output offset storage (OOS) technology is applied between the four-stage circuits of the high-speed latch comparator. The switches S1 to S4 are composed of two pairs of complementary CMOS switches. One end of the switches S1 to S4 is connected to the common mode level VCOM, and the other end of the switches S1 to S4 is respectively connected to the positive and negative input ends of the four-stage circuit of the comparator. The negative output terminal of the first latch comparison stage of the high-speed latch comparator is coupled to the positive input terminal of the second latch comparison stage by capacitor C1, the positive output terminal of the first latch comparison stage is coupled to the negative input terminal of the second latch comparison stage by capacitor C2, the negative output terminal of the second latch comparison stage is coupled to the positive input terminal of the third latch comparison stage by capacitor C3, the positive output terminal of the second latch comparison stage is coupled to the negative input terminal of the third latch comparison stage by capacitor C4, the negative output terminal of the third latch comparison stage is coupled to the negative input terminal of the output buffer stage by capacitor C5, and the positive output terminal of the third latch comparison stage is coupled to the positive input terminal of the output buffer stage by capacitor C6.
[0014] Beneficial effects: It can be seen from the technical solution provided by the present invention that the successive approximation analog-to-digital converter circuit provided by the present invention for analog in-memory calculation has advantages in power consumption and area compared with the traditional flash analog-to-digital converter under the high precision requirement of 8-bit codewords. Aiming at the demand for low-power integrated memory and computing chips, the present invention effectively controls the leakage of the current bias module and the bandgap reference module in the successive approximation analog-to-digital converter circuit, reduces the static power consumption, applies the output offset storage technology to the comparator module, and improves the linearity of the analog-to-digital converter. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.
[0016] Figure 1 A structural block diagram of an analog in-memory computing circuit in the prior art provided as background technology;
[0017] Figure 2 A schematic diagram of the structure of a matrix-vector multiplier circuit based on a crossbar array of flash memory devices in the prior art provided as background technology;
[0018] Figure 3 A schematic diagram of the structure of a flash analog-to-digital converter circuit in the prior art provided as background technology;
[0019] Figure 4 A schematic diagram of the overall structure of a successive approximation analog-to-digital converter (SAR ADC) circuit for analog in-memory computing proposed in an embodiment of the present invention;
[0020] Figure 5 A circuit schematic diagram of a current bias module with a startup circuit in a SAR ADC proposed in an embodiment of the present invention;
[0021] Figure 6 A circuit schematic diagram of a voltage-mode reference module controlled by an operational amplifier in a SAR ADC according to an embodiment of the present invention;
[0022] Figure 7 A circuit schematic diagram of a high-speed latch comparator module in a SAR ADC according to an embodiment of the present invention;
[0023] Figure 8 This is a schematic diagram of the working timing of the SAR ADC proposed in an embodiment of the present invention (when the output of the 4-bit ADC is 1010);
[0024] Fig. 9 This is a layout of the SAR ADC proposed in an embodiment of the present invention drawn under a 0.18um CMOS process. DETAILED DESCRIPTION
[0025] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the protection scope of the present invention.
[0026] Attached Figure 4The overall structure diagram of the SAR ADC circuit of the present invention also includes the circuit principle diagram of the 8-bit charge redistribution digital-to-analog converter, which is composed of capacitor arrays C0-C8, bridge capacitors Cc, and complementary CMOS switches S0-S8. Capacitors C1-C8 correspond to the lowest bit to the highest bit codeword of the 8-bit digital-to-analog conversion respectively, and capacitor C0 is a complementary capacitor, and the capacitance is the same as the capacitor C1 of the least significant bit (LSB). In order to reduce the area of the capacitor array, the bridge capacitor Cc is connected in series between the high 4-bit capacitor and the low 4-bit capacitor, and the capacitance ratio of the capacitors C0-C8 and Cc is 1:1:2:4:8:2:4:8:16:16 / 7. The lower plates of the capacitor arrays C0-C8 are connected to switches S0-S8 respectively, and the upper plates of the capacitor arrays, as the output end of the entire 8-bit charge redistribution digital-to-analog converter circuit, are connected to the positive input end of the latch comparator. Switches S0~S8 are controlled by the output signals Sample and B[7:0] of the successive approximation logic control circuit. The control signal Sample simultaneously regulates the switching between the analog input and the reference voltage of the entire analog-to-digital converter, as well as the resetting of the common-mode level in the upper plate of the capacitor array and the latch comparator.
[0027] Specifically, the upper 4-bit capacitors and the lower 4-bit capacitors of the capacitor array C0-C8 are connected in series using a bridge capacitor Cc. When the capacitance of capacitor Cc is 16 / 7 C0, the total capacitance of the capacitors C0-C4 plus the bridge capacitor Cc calculated by the series-parallel connection of the capacitors is exactly equal to the capacitance of capacitor C5, thus achieving the same effect as the traditional method of arranging capacitors exponentially. However, from the output end of the 8-bit charge redistribution DAC, the total capacitor load is 16 C0s, and the total capacitance of the capacitor array is 48.3 C0s. If the 8 capacitors are simply arranged exponentially, the total capacitor load and total capacitance will reach 256 C0s, greatly increasing the layout area.
[0028] Specifically, when the Sample signal is set high, the ADC first enters the sampling phase. At this time, the input node VP (V P ) and VN(V N ) is set to the common mode voltage V COM (V P =V N =V COM ). Set node VIN to the analog input voltage V IN After that, the control signal B[7:0] and Sample switch S0~S8 from V IN Switching to ground potential, through the AC coupling principle of capacitors C8~C0, the voltage at node VP is reduced by V IN , at this time there is V P =V COM -VIN .
[0029] After sampling is completed, the conversion phase begins and the Sample signal is pulled low. In this phase, the digital logic module generates a control signal to switch B[7] from ground potential to the reference voltage V provided by the bandgap reference. REF , then V P Increase 1 / 2V REF At this time, V P =V COM -V IN +1 / 2V REF The comparator will V P With V N Comparison, this is essentially V IN With 1 / 2V REF The output of the comparator is given to the digital logic module to determine the control of the capacitor switch in the next operation stage.
[0030] When V P >V N When V IN <1 / 2V REF In the next conversion phase, the digital logic module generates a control signal to change B[7] from V REF Switch to ground potential, switch B[6] from ground potential to V REF , which makes V P Reduced by 1 / 4V REF At this time, V P =V COM -V IN +1 / 4V REF .
[0031] When V P <V N When the digital logic module generates a control signal to switch B[6] from ground potential to V REF , while B[7] remains unchanged, thus V P Add another 1 / 4V REF At this time, V P = V COM -V IN +1 / 2V REF +1 / 4V REF .
[0032] After 8 conversion operations, the SAR ADC will determine B[7:0] to obtain the final digital output. The ADC working principle and timing described above are described in the attached Figure 8 A 4-bit SAR ADC with an output of 1010 is used as an example.
[0033] Attached Figure 5 The schematic diagram of the current bias circuit with the startup circuit described in the present invention. Specifically, two pairs of current mirrors of PMOS tubes M3 and M4 and NMOS tubes M5 and M6 are connected end to end to form a loop, and form a classic monotonic △VGS / R type self-bias circuit with resistor R1. Among them, the width-to-length ratio of PMOS tubes M3 and M4 is 1:1, and the width-to-length ratio of NMOS tubes M5 and M6 is 1:N (N>1). When the power is first turned on, the voltage on resistor R1 is approximately zero. At this time, the nonlinear current mirror below is approximately regarded as a linear current mirror. The current from the input branch to the output branch is amplified N times, and then copied to the input branch through the linear current mirror above. This cycle continues, which can be regarded as a positive feedback, and can eventually stabilize to a current bias that is approximately independent of the power supply voltage.
[0034] At the output end, VBIAS1 and VBIAS2 provide current bias for the reference circuit and buffer circuit respectively. Transistors M7 and M8 form an IV conversion current mirror from N-tube to P-tube, which copies the VBIAS2 current to the gate voltage on the PMOS tube, provides voltage bias VBIAS3 for the comparator, and isolates the influence of the latter stage on the former stage.
[0035] On this basis, the self-bias circuit has a robust startup circuit and has low leakage when the enable is zero.
[0036] Specifically, when the enable EN is 0, the transistor M2 is turned on, so that the gate voltage, drain voltage and gate voltage of the transistor M5 and M6 are clamped to the ground. One input of the NAND gate NAND1 via the inverters INV1 and INV2 is low level, and the other input of the NAND gate NAND1 via the drain of the transistor M5 via the inverter INV6 is high level, so the output of the NAND gate NAND1 is high level. Therefore, the output of the inverter INV4 is high level, which turns off the transistor M1 connected thereto. At the same time, since the output of the inverter INV5 is low level, the transistor M9 is turned off, and the transistor M10 is turned on, so the gate voltage, drain voltage and gate voltage of the transistor M4 and M3 are clamped to the power supply voltage VDD. At this time, both pairs of current mirrors are completely turned off, and no current flows in the branch.
[0037] When the enable EN starts from 0 to a high level, transistors M2 and M10 become cut off, and the shutoff effect on the current mirror disappears. At the same time, both inputs of the NAND gate NAND1 are turned high, so the output is low, which makes the transistor M1 start to conduct. At this time, the gate and drain of transistor M5 and the gate voltage of M6 are directly connected to the power supply voltage VDD. At the same time, since M9 is turned on, the gate and drain of transistor M4 and the gate voltage of M3 are clamped to the ground, and the entire bias circuit starts to power on quickly. When the bias circuit is powered on to a certain extent, the drain voltage of transistor M5 causes the output of inverter INV6 to reverse to a low level, and the output of the NAND gate NAND1 turns low again, which makes transistors M1 and M9 turn off. At this time, the work of the startup circuit is completed, and all startup circuits are isolated from the bias circuit.
[0038] Attached Figure 6 The schematic diagram of the voltage-mode reference circuit controlled by the operational amplifier described in the present invention is that based on the principle of the classic reference voltage source, two pairs of PMOS tubes M3, M4 and M5, M6, and a resistor R1 are used to form a common source and common gate structure, and another pair of PMOS tubes M1, M2 are added to enable the reference source. The two-stage operational amplifier used in the reference source also adds an NMOS tube M8 and a PMOS tube M9 to enable the two stages of the operational amplifier respectively.
[0039] Attached Figure 7 The schematic diagram of the high-speed latch comparator circuit of the present invention is composed of three latch comparison stages and one output buffer stage. The three latch comparison stages are static latch circuits with the same structure. Their function is to compare the analog voltage at the positive end of the input comparator with the common mode level at the negative end, and then output the logic level. The output buffer stage circuit is a simple differential op amp, which provides sufficient gain for the comparator and outputs a stable logic level.
[0040] Specifically, the output offset storage (OOS) technology is applied between the four-stage circuits of the high-speed latch comparator, and the switches S1-S4 are each composed of two pairs of complementary CMOS switches, one end of the switches S1-S4 is connected to the common mode level VCOM, and the other ends of the switches S1-S4 are respectively connected to the positive and negative input ends of the four-stage circuit of the comparator. The negative output end of the first latch comparison stage of the high-speed latch comparator is coupled with the positive input end of the second latch comparison stage by a capacitor C1, the positive output end of the first latch comparison stage is coupled with the negative input end of the second latch comparison stage by a capacitor C2, the negative output end of the second latch comparison stage is coupled with the positive input end of the third latch comparison stage by a capacitor C3, the positive output end of the second latch comparison stage is coupled with the negative input end of the third latch comparison stage by a capacitor C4, the negative output end of the third latch comparison stage is coupled with the negative input end of the output buffer stage by a capacitor C5, and the positive output end of the third latch comparison stage is coupled with the positive input end of the output buffer stage by a capacitor C6. At the beginning of the ADC sampling phase, the digital logic module sends a control signal to close switches S1 to S4 at the same time, ensuring that the input and output potentials of each level of the comparator are reset to the VCOM common mode level. Before the end of the sampling phase, switches S4 to S1 are opened one by one.
[0041] Attached Fig. 9 This is the layout of the SAR ADC proposed in the present invention drawn under the 0.18um CMOS process. The layout area is about 464um×130um. After extracting the parasitic parameters, the simulation shows that under PVT conditions (process angles tt, ff, ss, power supply voltage VDD=1.62V, 1.8V, 1.98V, temperature -40℃, 27℃, 125℃), the average ADC static current is less than 10nA, the average operating current is less than 50uA, the maximum absolute error is less than 7mV, and the ±INL is less than 1.5LSB.
[0042] Aiming at the demand for low-power storage and computing integrated chips, the present invention effectively controls the leakage of the current bias module and the bandgap reference module in the successive approximation analog-to-digital converter circuit, reduces the static power consumption, applies the output offset storage technology to the comparator module, and improves the linearity of the analog-to-digital converter.
[0043] Parts of the present invention that are not disclosed in detail belong to the common knowledge in the art.
[0044] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.
Claims
1. A successive approximation analog-to-digital converter circuit for analog in-memory computing, characterized in that: The circuit includes the following modules: An 8-bit charge redistribution type digital-to-analog converter circuit, a current bias circuit with a startup circuit, a voltage-mode reference circuit controlled by an operational amplifier, a high-speed latch comparator circuit and a successive approximation logic control module for implementing a binary search algorithm; the current bias circuit is connected to the voltage-mode reference circuit, a buffer circuit attached to the voltage-mode reference circuit and a high-speed latch comparator circuit to provide current or voltage bias; the digital-to-analog converter circuit controls the digital input by the successive approximation logic control module, and uses a high-speed latch comparator circuit to compare the analog input and the reference voltage, and the basic comparison result is fed back to the digital input of the digital-to-analog converter circuit for adjustment, and finally realizes the successive approximation analog-to-digital conversion function in the form of a binary search algorithm; the high-speed latch comparator circuit is composed of three latch comparison stages and one output buffer stage, and the three latch comparison stages are static latch circuits with the same structure, and their function is to compare and determine the analog voltage at the positive end of the input comparator with the common mode level at the negative end, and then output the logic level; The output buffer stage circuit is a simple differential op amp, which provides sufficient gain and outputs a stable logic level for the high-speed latch comparator circuit; The output offset storage OOS technology is applied between the four-stage circuits of the high-speed latch comparator, and the switches S1-S4 are each composed of two pairs of complementary CMOS switches, one end of the switches S1-S4 are all connected to the common mode level VCOM, and the other ends of the switches S1-S4 are respectively connected to the positive and negative input ends of the four-stage circuit of the comparator; the negative output end of the first latch comparison stage of the high-speed latch comparator is coupled with the positive input end of the second latch comparison stage by a capacitor C1, the positive output end of the first latch comparison stage is coupled with the negative input end of the second latch comparison stage by a capacitor C2, the negative output end of the second latch comparison stage is coupled with the positive input end of the third latch comparison stage by a capacitor C3, the positive output end of the second latch comparison stage is coupled with the negative input end of the third latch comparison stage by a capacitor C4, the negative output end of the third latch comparison stage is coupled with the negative input end of the output buffer stage by a capacitor C5, and the positive output end of the third latch comparison stage is coupled with the positive input end of the output buffer stage by a capacitor C6.
2. The successive approximation analog-to-digital converter circuit for analog in-memory computing according to claim 1, characterized in that: The 8-bit charge redistribution digital-to-analog converter circuit is composed of a capacitor array C0-C8, a bridge capacitor Cc, and complementary CMOS switches S0-S8; the capacitors C1-C8 correspond to the lowest bit to the highest bit codeword of the 8-bit digital-to-analog conversion respectively, the capacitor C0 is a complementary capacitor, and the capacitance is the same as the capacitor C1 of the least significant bit LSB; in order to reduce the area of the capacitor array, the bridge capacitor Cc is connected in series between the upper 4-bit capacitor and the lower 4-bit capacitor, and the capacitance ratio of the capacitors C0-C8 and Cc is 1:1:2:4:8:2:4:8:16:16 / 7; The lower plates of the capacitor array C0-C8 are connected to switches S0-S8 respectively, and the upper plate of the capacitor array, as the output end of the entire 8-bit charge redistribution digital-to-analog converter circuit, is connected to the positive input end of the latch comparator; the switches S0-S8 are controlled by the output signals Sample and B[7:0] of the successive approximation logic control circuit, and the control signal Sample simultaneously regulates the switching between the analog input and the reference voltage of the entire analog-to-digital converter, as well as the resetting of the common mode level in the upper plate of the capacitor array and the latch comparator.
3. The successive approximation analog-to-digital converter circuit for analog in-memory computing according to claim 1, characterized in that: The current bias circuit with startup circuit is composed of two pairs of current mirrors forming a loop and a resistor to form a classic monotonic △V GS A / R type self-bias circuit, wherein a PMOS tube current mirror and an NMOS tube current mirror respectively provide current bias for a reference circuit in an analog-to-digital converter and a buffer circuit after the reference circuit, and then an NMOS tube and a PMOS tube are used to copy the current to the gate voltage on the PMOS tube, thereby providing a voltage bias for a high-speed comparator in the analog-to-digital converter and isolating the influence of the subsequent stage; on this basis, the self-bias circuit has a set of robust startup circuits and has low leakage when the enable is zero.
4. The successive approximation analog-to-digital converter circuit for analog in-memory computing according to claim 1, characterized in that: The voltage-mode reference circuit controlled by the operational amplifier uses two pairs of PMOS tubes and a resistor to form a common source and common gate structure, and an additional pair of PMOS tubes is added to enable the reference source. The two-stage operational amplifier used in the reference source also adds an NMOS tube and a PMOS tube to enable the two stages of the operational amplifier respectively.
Citation Information
Patent Citations
Cascade comparator circuit
CN106603048A
Low-power-consumption successive approximation type analog-to-digital converter and analog-to-digital conversion method
CN111181563A
Successive approximation register analog-to-digital converter
US20150162926A1
Dual-use comparator / op amp for use as both a successive-approximation ADC and DAC
US7741981B1
