Zhongzhen Tong
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Zhongzhen Tong has a total of 95 co-authors. Here are the top co-authors by number of shared publications:
- Weisheng Zhao
National Key Laboratory of Spintronics, Hangzhou International Innovation Institute, Beihang University 1 , Hangzhou 311115
9 shared publications - Xiaoyang Lin
State Key Laboratory of Spintronics, Hangzhou International Innovation Institute, Beihang University 1 , Hangzhou 311115
9 shared publications - Xiulong Wu
Anhui University
8 shared publications - Zhiting Lin
Anhui University
8 shared publications - Zhaohao Wang
School of Integrated Circuit Science and Engineering Beihang University Beijing China
6 shared publications Chenghang Li
State Key Laboratory of Spintronics, Hangzhou International Innovation Institute, and the School of Integrated Circuit Science and Engineering, Beihang University, Hangzhou, China
5 shared publications
Latest Publications
- 21 September 2026
- Published by AIP Publishing in Applied Physics Letters
- Vol. 129  (12)
Total ionizing dose and single-event effect of spin–orbit torque magnetic tunnel junctions with different bias conditions
Spin–orbit torque magnetic tunnel junctions (SOT-MTJs) have attracted interest as potential candidates for radiation-tolerant memory applications because of their high speed, low power consumption, and high endurance. Although SOT-MTJs utilize spin for data storage, operations are performed by applying different electrical biases. In conventional silicon-based devices, the applied electric field typically accelerates radiation-induced defect accumulation and dielectric breakdown, and the dynamic radiation stability of SOT-MTJs under these conditions remains largely unexplored. Therefore, we investigated the total ionizing dose and single-event effects of SOT-MTJs under electrical bias using 10 keV x-ray and 1673 MeV Ta-ion irradiation, respectively. Under various bias conditions and dose rates, the tunnel magnetoresistance (TMR) and critical switching current (Ic) of the SOT-MTJs remained highly stable even at a total dose of 1.5 Mrad(Si), with TMR fluctuations of less than 2%. Devices subjected to heavy-ion irradiation exhibited no abnormal switching or breakdown in either the anti-parallel (AP) or parallel (P) state. Furthermore, the average TMR variation in the unbiased device arrays was approximately 1%, while the average Ic remained nearly unchanged. These results demonstrate the radiation stability of the fabricated 320 × 1100 nm2 in-plane SOT-MTJ test vehicles under the investigated total ionizing-dose and heavy-ion irradiation conditions, while providing a device-level reference for future radiation studies of scaled SOT-MRAM technologies.
- 21 September 2026
- Published by AIP Publishing in Applied Physics Letters
- Vol. 129  (12)
Total ionizing dose and single-event effect of spin–orbit torque magnetic tunnel junctions with different bias conditions
Spin–orbit torque magnetic tunnel junctions (SOT-MTJs) have attracted interest as potential candidates for radiation-tolerant memory applications because of their high speed, low power consumption, and high endurance. Although SOT-MTJs utilize spin for data storage, operations are performed by applying different electrical biases. In conventional silicon-based devices, the applied electric field typically accelerates radiation-induced defect accumulation and dielectric breakdown, and the dynamic radiation stability of SOT-MTJs under these conditions remains largely unexplored. Therefore, we investigated the total ionizing dose and single-event effects of SOT-MTJs under electrical bias using 10 keV x-ray and 1673 MeV Ta-ion irradiation, respectively. Under various bias conditions and dose rates, the tunnel magnetoresistance (TMR) and critical switching current (Ic) of the SOT-MTJs remained highly stable even at a total dose of 1.5 Mrad(Si), with TMR fluctuations of less than 2%. Devices subjected to heavy-ion irradiation exhibited no abnormal switching or breakdown in either the anti-parallel (AP) or parallel (P) state. Furthermore, the average TMR variation in the unbiased device arrays was approximately 1%, while the average Ic remained nearly unchanged. These results demonstrate the radiation stability of the fabricated 320 × 1100 nm2 in-plane SOT-MTJ test vehicles under the investigated total ionizing-dose and heavy-ion irradiation conditions, while providing a device-level reference for future radiation studies of scaled SOT-MRAM technologies.
- 6 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
- Vol. 34  (3) , 904-917
A CIM Macro Embedded With Sign Operations for Parallel Signed Multibit Multiplication-and-Accumulation Using Hybrid Cell Array
Multiplication is a fundamental operation in neural network models. However, signed multibit multiplication and accumulation (MAC) pose significant challenges, primarily due to the complexities involved in handling sign bits. Relying solely on auxiliary modules for sign bit operations in computing-in-memory (CIM) arrays increases time and area overhead. To address this, we propose a CIM signed MAC architecture utilizing a hybrid cell array. Our proposed strategy relies on the partial structure within an 11 T cell to complete sign operations, rather than relying on additional modules; it achieves a balance between area consumption and computational efficiency. Our strategy decomposes the discharge that represents the operation value according to weights and distributes it across different bit lines. Therefore, under the limited bitline voltage margin, the discharge value of the least significant bit does not sharply decrease as the number of operation bits increases, which ensures the accuracy of the MAC operation. Since all the cells in the array can simultaneously perform the corresponding sign or numerical operations, and numerical accumulation is reflected in the analog voltage physical quantities, each group of MAC operations can be performed with high parallelism. Simulations using a 28-nm CMOS process demonstrate that the proposed circuit achieves high linearity, such that the absolute value of the integral nonlinearity (INL) does not exceed 1.12. The MAC operation achieves an energy efficiency of 26.9 TOPS/W at a frequency of 417 MHz, with 4-bit input, 4-bit weight, and 4-bit output. Our proposed strategy demonstrates an advantage in figure-of-merit (FOM) and network data accuracy.
- 6 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
- Vol. 34  (3) , 904-917
A CIM Macro Embedded With Sign Operations for Parallel Signed Multibit Multiplication-and-Accumulation Using Hybrid Cell Array
Multiplication is a fundamental operation in neural network models. However, signed multibit multiplication and accumulation (MAC) pose significant challenges, primarily due to the complexities involved in handling sign bits. Relying solely on auxiliary modules for sign bit operations in computing-in-memory (CIM) arrays increases time and area overhead. To address this, we propose a CIM signed MAC architecture utilizing a hybrid cell array. Our proposed strategy relies on the partial structure within an 11 T cell to complete sign operations, rather than relying on additional modules; it achieves a balance between area consumption and computational efficiency. Our strategy decomposes the discharge that represents the operation value according to weights and distributes it across different bit lines. Therefore, under the limited bitline voltage margin, the discharge value of the least significant bit does not sharply decrease as the number of operation bits increases, which ensures the accuracy of the MAC operation. Since all the cells in the array can simultaneously perform the corresponding sign or numerical operations, and numerical accumulation is reflected in the analog voltage physical quantities, each group of MAC operations can be performed with high parallelism. Simulations using a 28-nm CMOS process demonstrate that the proposed circuit achieves high linearity, such that the absolute value of the integral nonlinearity (INL) does not exceed 1.12. The MAC operation achieves an energy efficiency of 26.9 TOPS/W at a frequency of 417 MHz, with 4-bit input, 4-bit weight, and 4-bit output. Our proposed strategy demonstrates an advantage in figure-of-merit (FOM) and network data accuracy.
- 2 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. PP  (15498328) , 1-14
BaM-CIM: A High Throughput Booth Algorithm-Based In-MRAM Computing Macro Using Hybrid VGSOT-MTJ/GAA-CNTFET
As artificial intelligence (AI) and computational models grow in scale, the demand for computational power and storage has significantly increased. The computing-in-memory (CIM) architecture addresses this challenge by performing computations directly within the memory array, reducing data transfer between the processor and memory. This paper introduces a Booth algorithm-based In-MRAM computing architecture (BaM-CIM) using a hybrid voltage-gated spin-orbit torque MTJ (VGSOT-MTJ) and gate-all-around carbon nanotube field-effect transistors (GAA-CNTFETs) for efficient multiply-and-accumulate (MAC) computing. The key contributions of BaM-CIM are as follows: 1) A Voltage divider reference (VDR) cell is proposed, which enables read operations using only a 2T1M cell structure. Compared to complementary read cells, the VDR reduces the area by half and achieves robust data sensing without requiring precharge/discharge operations. 2) The BaM-CIM circuit is proposed to complete 8b-W/8b-IN/21b-OUT computations in only two cycles (1.6 ns), reducing the number of cycles by 75% compared to single-bit input serial operations and by 50% compared to two-bit serial operations. 3) A three-input 8b Booth computing adder (BCA), along with Modified computing shift adder (MCSA) and Modified computing post adder (MCPA), which can achieve higher energy efficiency. BaM-CIM with 128 Kb is simulated, achieving throughput and energy efficiency of 0.93 TOPS and 258.4 TOPS/W, respectively, at a 0.6 V supply voltage and 1.28 TOPS and 169.5 TOPS/W, respectively, at a 0.8 V supply voltage with 8b-IN, 8b-W, and 21b-OUT.
- 2 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. PP  (15498328) , 1-14
BaM-CIM: A High Throughput Booth Algorithm-Based In-MRAM Computing Macro Using Hybrid VGSOT-MTJ/GAA-CNTFET
As artificial intelligence (AI) and computational models grow in scale, the demand for computational power and storage has significantly increased. The computing-in-memory (CIM) architecture addresses this challenge by performing computations directly within the memory array, reducing data transfer between the processor and memory. This paper introduces a Booth algorithm-based In-MRAM computing architecture (BaM-CIM) using a hybrid voltage-gated spin-orbit torque MTJ (VGSOT-MTJ) and gate-all-around carbon nanotube field-effect transistors (GAA-CNTFETs) for efficient multiply-and-accumulate (MAC) computing. The key contributions of BaM-CIM are as follows: 1) A Voltage divider reference (VDR) cell is proposed, which enables read operations using only a 2T1M cell structure. Compared to complementary read cells, the VDR reduces the area by half and achieves robust data sensing without requiring precharge/discharge operations. 2) The BaM-CIM circuit is proposed to complete 8b-W/8b-IN/21b-OUT computations in only two cycles (1.6 ns), reducing the number of cycles by 75% compared to single-bit input serial operations and by 50% compared to two-bit serial operations. 3) A three-input 8b Booth computing adder (BCA), along with Modified computing shift adder (MCSA) and Modified computing post adder (MCPA), which can achieve higher energy efficiency. BaM-CIM with 128 Kb is simulated, achieving throughput and energy efficiency of 0.93 TOPS and 258.4 TOPS/W, respectively, at a 0.6 V supply voltage and 1.28 TOPS and 169.5 TOPS/W, respectively, at a 0.8 V supply voltage with 8b-IN, 8b-W, and 21b-OUT.
- 11 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 73  (2) , 875-885
An FD-SOI-Based Compact In-Pixel Computing Architecture Enabling Real-Time Feature Extraction
To empower resource-limited edge devices in artificial intelligence (AI) and Internet of Things (IoT) applications, it is essential to overcome challenges posed by restricted area resources and the high latency demands of transmitting and processing substantial sensory data. In-pixel computing addresses these challenges effectively, and the Fully Depleted Silicon-On-Insulator (FD-SOI)-based pixel, which relies on an FD-SOI transistor whose current is made photosensitive to light by applying a negative back-gate voltage, shows significant potential with its compact structure and in-situ computation capability. In this paper, for the first time, we present an FD-SOI-based chip-level architecture for in-pixel computing. Our design implements programmable, massively parallel convolution with low latency using a pulse-width modulation (PWM) input encoding scheme. Furthermore, the proposed compact 1P1T (1 Phototransistor 1 Transistor) pixel design, integrated with an improved single-slope analog-to-digital converter (SS ADC), greatly enhances area efficiency. Validated through simulation in a 22nm FD-SOI process, the design achieves 990 frames/s under typical outdoor illumination conditions, with the figure of merit (FoM) of 16.86 pJ/pixel/frame. In addition, the proposed architecture has been evaluated on hand gesture recognition (5697 training and 633 validation images across six categories), achieving an accuracy of 97.48%. The results demonstrate that, compared to state-of-the-art designs, our approach achieves a$6\times $ improvement in in-pixel convolution speed and a$7.3\times $ reduction in area overhead.- 11 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 73  (2) , 875-885
An FD-SOI-Based Compact In-Pixel Computing Architecture Enabling Real-Time Feature Extraction
To empower resource-limited edge devices in artificial intelligence (AI) and Internet of Things (IoT) applications, it is essential to overcome challenges posed by restricted area resources and the high latency demands of transmitting and processing substantial sensory data. In-pixel computing addresses these challenges effectively, and the Fully Depleted Silicon-On-Insulator (FD-SOI)-based pixel, which relies on an FD-SOI transistor whose current is made photosensitive to light by applying a negative back-gate voltage, shows significant potential with its compact structure and in-situ computation capability. In this paper, for the first time, we present an FD-SOI-based chip-level architecture for in-pixel computing. Our design implements programmable, massively parallel convolution with low latency using a pulse-width modulation (PWM) input encoding scheme. Furthermore, the proposed compact 1P1T (1 Phototransistor 1 Transistor) pixel design, integrated with an improved single-slope analog-to-digital converter (SS ADC), greatly enhances area efficiency. Validated through simulation in a 22nm FD-SOI process, the design achieves 990 frames/s under typical outdoor illumination conditions, with the figure of merit (FoM) of 16.86 pJ/pixel/frame. In addition, the proposed architecture has been evaluated on hand gesture recognition (5697 training and 633 validation images across six categories), achieving an accuracy of 97.48%. The results demonstrate that, compared to state-of-the-art designs, our approach achieves a$6\times $ improvement in in-pixel convolution speed and a$7.3\times $ reduction in area overhead.- Open Access
- 22 August 2025
- Published by American Association for the Advancement of Science (AAAS) in Science Advances
- Vol. 11  (34) , eadw0024
Large-scale complementary carbon nanotube integrated circuits for harsh radiation environments
Silicon-based integrated circuits operating in radiation environments require additional and complex hardening configurations, leading to performance lags compared to the International Roadmap for Devices and Systems. Carbon nanotubes (CNTs), with their ultrastrong chemical bonds and nanoscale dimensions, offer substantial potential for high-performance, radiation-tolerant electronics. However, the challenges associated with radiation-tolerant fabrication processes have hindered the development of macroelectronics using complementary CNT transistors (CNTFETs). In this study, we successfully fabricated radiation-tolerant, highly symmetric, and uniform CMOS building blocks, implementing various logic gates (inverters, NAND, and XOR gates) and ring oscillators (ROs) with 5, 11, and 501 stages. After irradiation up to 6 Mrad(Si), all devices maintained rail-to-rail outputs, and notably, the 501-stage RO, comprising 1004 CNTFETs, showed minimal delay variation (10.3 ± 0.8 ns). This work demonstrates the radiation-tolerant of large-scale CNTFETs, paving the way for their potential replacement of silicon-based FETs in radiation-heavy environments.
- 22 August 2025
- Published by American Association for the Advancement of Science (AAAS) in Science Advances
- Vol. 11  (34) , eadw0024
Open AccessLarge-scale complementary carbon nanotube integrated circuits for harsh radiation environments
Silicon-based integrated circuits operating in radiation environments require additional and complex hardening configurations, leading to performance lags compared to the International Roadmap for Devices and Systems. Carbon nanotubes (CNTs), with their ultrastrong chemical bonds and nanoscale dimensions, offer substantial potential for high-performance, radiation-tolerant electronics. However, the challenges associated with radiation-tolerant fabrication processes have hindered the development of macroelectronics using complementary CNT transistors (CNTFETs). In this study, we successfully fabricated radiation-tolerant, highly symmetric, and uniform CMOS building blocks, implementing various logic gates (inverters, NAND, and XOR gates) and ring oscillators (ROs) with 5, 11, and 501 stages. After irradiation up to 6 Mrad(Si), all devices maintained rail-to-rail outputs, and notably, the 501-stage RO, comprising 1004 CNTFETs, showed minimal delay variation (10.3 ± 0.8 ns). This work demonstrates the radiation-tolerant of large-scale CNTFETs, paving the way for their potential replacement of silicon-based FETs in radiation-heavy environments.
- Open Access
- 1 July 2025
- Published by AIP Publishing in AIP Advances
- Vol. 15  (7)
Phase-change controlled spin–orbit torque magnetic tunnel junction for in-memory sensing and computing
We present a device integrated thermal and optical induced phase change materials vanadium dioxide and spin–orbit torque magnetic tunnel junction. This device showcases the ability to perform in-memory sensing and computing under thermal and optical stimulus, manifesting multiple logic functions in response to different input signals. Furthermore, a circuit simulation model based on the experimental results is built, which can implement 16 types of reconfigurable Boolean logic gates. On this basis, a 1 kb half-adder Magnetoresistive Random Access Memory (MRAM) array is constructed that significantly enhancing the computational throughput.
- 1 July 2025
- Published by AIP Publishing in AIP Advances
- Vol. 15  (7)
Open AccessPhase-change controlled spin–orbit torque magnetic tunnel junction for in-memory sensing and computing
We present a device integrated thermal and optical induced phase change materials vanadium dioxide and spin–orbit torque magnetic tunnel junction. This device showcases the ability to perform in-memory sensing and computing under thermal and optical stimulus, manifesting multiple logic functions in response to different input signals. Furthermore, a circuit simulation model based on the experimental results is built, which can implement 16 types of reconfigurable Boolean logic gates. On this basis, a 1 kb half-adder Magnetoresistive Random Access Memory (MRAM) array is constructed that significantly enhancing the computational throughput.
- 12 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 72  (11) , 6988-7001
A Self-Decryption Pass Transistor Logic-Based In-MRAM Computing Macro Using Hybrid VGSOT-MTJ/GAA-CNTFET
Spintronic devices and gate-all-around carbon nanotube field-effect-transistors (GAA-CNTFETs)-based computing in-memory architecture are competitive candidates for applications in battery-powered tiny artificial intelligence (AI) edge devices. Meanwhile, data encryption and decryption are also necessary to protect AI model weights and the customized data used to guarantee neural network (NN) inference accuracy. In this study, we propose a self-decryption pass transistor logic (PTL)-based in-MRAM computing macro (SP-CIM) that utilizes hybrid voltage-gated spin-orbit torque magnetic tunnel junctions (VGSOT-MTJ)/GAA-CNTFET. The proposed SP-CIM macro enables simultaneous data access, decryption, and full-accuracy multiply-and-accumulate (MAC) operations using the newly introduced voltage-divider self-decryption cell, without the need for additional decryption logic. Compared to existing in-memory decryption strategies, this design reduces energy consumption by 45.7% and decreases decryption delay by 87.2%. To enhance area efficiency and reduce computing latency, we propose a PTL-based multiplication cell that achieves full-accuracy local 2b-IN $\times 2$ b-W operations with only 20 transistors (20T). Additionally, novel PTL-based full-swing output half adders (10T-HA) and full adders (14T-FA) are proposed to construct the local adder tree, achieving reductions of 31.8%, 76.4%, and 41.4% in energy, delay, and area, respectively, compared to conventional adder trees in CIM macros. Simulations of the 288 kb SP-CIM macro demonstrated throughput and energy efficiency of 2.25 TOPS and 226.6 TOPS/W, respectively, at a 0.6 V supply voltage, and 2.97 TOPS and 154.1 TOPS/W, respectively, at a 0.8 V supply voltage, with 8b-IN, 8b-W, and 24b-OUT.
- 12 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 72  (11) , 6988-7001
A Self-Decryption Pass Transistor Logic-Based In-MRAM Computing Macro Using Hybrid VGSOT-MTJ/GAA-CNTFET
Spintronic devices and gate-all-around carbon nanotube field-effect-transistors (GAA-CNTFETs)-based computing in-memory architecture are competitive candidates for applications in battery-powered tiny artificial intelligence (AI) edge devices. Meanwhile, data encryption and decryption are also necessary to protect AI model weights and the customized data used to guarantee neural network (NN) inference accuracy. In this study, we propose a self-decryption pass transistor logic (PTL)-based in-MRAM computing macro (SP-CIM) that utilizes hybrid voltage-gated spin-orbit torque magnetic tunnel junctions (VGSOT-MTJ)/GAA-CNTFET. The proposed SP-CIM macro enables simultaneous data access, decryption, and full-accuracy multiply-and-accumulate (MAC) operations using the newly introduced voltage-divider self-decryption cell, without the need for additional decryption logic. Compared to existing in-memory decryption strategies, this design reduces energy consumption by 45.7% and decreases decryption delay by 87.2%. To enhance area efficiency and reduce computing latency, we propose a PTL-based multiplication cell that achieves full-accuracy local 2b-IN $\times 2$ b-W operations with only 20 transistors (20T). Additionally, novel PTL-based full-swing output half adders (10T-HA) and full adders (14T-FA) are proposed to construct the local adder tree, achieving reductions of 31.8%, 76.4%, and 41.4% in energy, delay, and area, respectively, compared to conventional adder trees in CIM macros. Simulations of the 288 kb SP-CIM macro demonstrated throughput and energy efficiency of 2.25 TOPS and 226.6 TOPS/W, respectively, at a 0.6 V supply voltage, and 2.97 TOPS and 154.1 TOPS/W, respectively, at a 0.8 V supply voltage, with 8b-IN, 8b-W, and 24b-OUT.
- 16 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems II: Express Briefs
- Vol. 72  (1) , 178-182
A Novel Radiation-Hardened, Speed and Power Optimized Nonvolatile Latch for Aerospace Applications
As process nodes shrink to deep nanometer scales, the heightened sensitivity of digital circuits to radiation and the dramatic rise in leakage power have become pressing concerns. Magnetic tunnel junction (MTJ) possesses intrinsic radiation resistance and nonvolatility and can be integrated with CMOS processes. Therefore, designing MTJ-based radiation-hardened nonvolatile storage elements is a promising solution to address these issues. In this brief, we propose a novel MTJ-based radiation-hardened, speed and power optimized nonvolatile (RH-SPO) latch and compares it with existing designs to evaluate its performance. Using a 28 nm CMOS process, simulation results confirm that the proposed RH-SPO latch offers moderate radiation resistance and high robustness performance with backup and restore operations. Additionally, compared to state-of-the-art design, named M-8C, it can save up to 50% in area overhead, reduce transmission-restore power consumption and restore time by 98% and 75%, respectively.
- 16 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems II: Express Briefs
- Vol. 72  (1) , 178-182
A Novel Radiation-Hardened, Speed and Power Optimized Nonvolatile Latch for Aerospace Applications
As process nodes shrink to deep nanometer scales, the heightened sensitivity of digital circuits to radiation and the dramatic rise in leakage power have become pressing concerns. Magnetic tunnel junction (MTJ) possesses intrinsic radiation resistance and nonvolatility and can be integrated with CMOS processes. Therefore, designing MTJ-based radiation-hardened nonvolatile storage elements is a promising solution to address these issues. In this brief, we propose a novel MTJ-based radiation-hardened, speed and power optimized nonvolatile (RH-SPO) latch and compares it with existing designs to evaluate its performance. Using a 28 nm CMOS process, simulation results confirm that the proposed RH-SPO latch offers moderate radiation resistance and high robustness performance with backup and restore operations. Additionally, compared to state-of-the-art design, named M-8C, it can save up to 50% in area overhead, reduce transmission-restore power consumption and restore time by 98% and 75%, respectively.
- 13 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems II: Express Briefs
- Vol. 72  (1) , 173-177
In-MRAM Computing Based on Complementary-Sensing Time-Based Readout Circuit Using Hybrid VGSOT-MTJ/GAA-CNTFET
Gate-all-around carbon nanotube field-effect-transistors (GAA-CNTFETs) and voltage-gated spin-orbit torque magnetic tunnel junctions (VGSOT-MTJs) are expected to realize significant savings in energy consumption and computing delay compared to the existing silicon-based FinFETs. This brief proposes an in-MRAM computing macro based on a newly developed complementary-sensing time-based readout circuit (CSTRC) to accelerate binary neural networks (BNNs). An 8 kb MRAM was simulated using both GAA-CNTFET/VGSOT-MTJ and 14 nm FinFET/VGSOT-MTJ technologies to validate the effectiveness of the proposed design. The proposed CSTRC can achieve read operations and binary multiply-and-accumulate (BMAC) without additional peripheral circuits and achieve a notable decrease in the read bit error rate and column-level conditional row error rate by 1–5 and 1–13 orders of magnitude, respectively, compared to those reported previously. Moreover, under the GAA-CNTFET/VGSOT-MTJ process, the read energy consumption and delay were reduced by 59.1–78.9% and 23.9–29.7%, respectively; the BMAC energy efficiency and throughput were 10231 1-b TOPS/W and 1.8 TOPS, respectively increased by 2.9 and 1.27 times at 0.8 V supply voltage when comparing to its 14-nm FinFET /VGSOT-MTJ counterparts.
- 13 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems II: Express Briefs
- Vol. 72  (1) , 173-177
In-MRAM Computing Based on Complementary-Sensing Time-Based Readout Circuit Using Hybrid VGSOT-MTJ/GAA-CNTFET
Gate-all-around carbon nanotube field-effect-transistors (GAA-CNTFETs) and voltage-gated spin-orbit torque magnetic tunnel junctions (VGSOT-MTJs) are expected to realize significant savings in energy consumption and computing delay compared to the existing silicon-based FinFETs. This brief proposes an in-MRAM computing macro based on a newly developed complementary-sensing time-based readout circuit (CSTRC) to accelerate binary neural networks (BNNs). An 8 kb MRAM was simulated using both GAA-CNTFET/VGSOT-MTJ and 14 nm FinFET/VGSOT-MTJ technologies to validate the effectiveness of the proposed design. The proposed CSTRC can achieve read operations and binary multiply-and-accumulate (BMAC) without additional peripheral circuits and achieve a notable decrease in the read bit error rate and column-level conditional row error rate by 1–5 and 1–13 orders of magnitude, respectively, compared to those reported previously. Moreover, under the GAA-CNTFET/VGSOT-MTJ process, the read energy consumption and delay were reduced by 59.1–78.9% and 23.9–29.7%, respectively; the BMAC energy efficiency and throughput were 10231 1-b TOPS/W and 1.8 TOPS, respectively increased by 2.9 and 1.27 times at 0.8 V supply voltage when comparing to its 14-nm FinFET /VGSOT-MTJ counterparts.
- 9 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 71  (12) , 6114-6127
BSTCIM: A Balanced Symmetry Ternary Fully Digital In-MRAM Computing Macro for Energy Efficiency Neural Network
Silicon-based traditional binary computing in-memory (TBCIM) architectures are approaching their energy efficiency and throughput limits owing to challenges facing Moore’s Law. Thus, it is essential to explore architecture based on novel devices and computing paradigms to fulfill data-centric applications, such as artificial intelligence. In this paper, we propose a balanced symmetry ternary (BST) fully digital in-MRAM computing macro (BSTCIM) using hybrid voltage-gated spin-orbit torque magnetic tunnel junctions (VGSOT-MTJ) and gate-all-around carbon nanotube field-effect-transistors (GAA-CNTFET) technology. The overall computing is based on the highest efficiency multi-bit ternary system. BSTCIM includes a ternary dot product (TDP) unit with 4 GAA-CNTFETs and 2 VGSOT-MTJs achieving TDP operation without complex logic circuits. The multi-bit ternary multiply-and-accumulate (MAC) operation is realized through the proposed ternary adder tree and ternary post adder which accumulate TDP results within the digital domain enabling high accuracy neural network inference. Furthermore, due to the advantages of BST, ternary signed MAC is more easily performed compared to TBCIM macros that adapt 2’s complement or separate signed bit calculations. BSTCIM with 288 kb is simulated, achieving throughput and energy efficiency of 0.72 TOPS and 54.5 TOPS/W, respectively, at a 0.6 V supply voltage and 1.15 TOPS and 33.7 TOPS/W, respectively at a 0.8 V supply voltage with 8b-IN, 8b-W, and 20b-OUT. Moreover, the figure-of-merit for BSTCIM is 1.13–33.6 times higher than that of existing CIM macros.
- 9 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 71  (12) , 6114-6127
BSTCIM: A Balanced Symmetry Ternary Fully Digital In-MRAM Computing Macro for Energy Efficiency Neural Network
Silicon-based traditional binary computing in-memory (TBCIM) architectures are approaching their energy efficiency and throughput limits owing to challenges facing Moore’s Law. Thus, it is essential to explore architecture based on novel devices and computing paradigms to fulfill data-centric applications, such as artificial intelligence. In this paper, we propose a balanced symmetry ternary (BST) fully digital in-MRAM computing macro (BSTCIM) using hybrid voltage-gated spin-orbit torque magnetic tunnel junctions (VGSOT-MTJ) and gate-all-around carbon nanotube field-effect-transistors (GAA-CNTFET) technology. The overall computing is based on the highest efficiency multi-bit ternary system. BSTCIM includes a ternary dot product (TDP) unit with 4 GAA-CNTFETs and 2 VGSOT-MTJs achieving TDP operation without complex logic circuits. The multi-bit ternary multiply-and-accumulate (MAC) operation is realized through the proposed ternary adder tree and ternary post adder which accumulate TDP results within the digital domain enabling high accuracy neural network inference. Furthermore, due to the advantages of BST, ternary signed MAC is more easily performed compared to TBCIM macros that adapt 2’s complement or separate signed bit calculations. BSTCIM with 288 kb is simulated, achieving throughput and energy efficiency of 0.72 TOPS and 54.5 TOPS/W, respectively, at a 0.6 V supply voltage and 1.15 TOPS and 33.7 TOPS/W, respectively at a 0.8 V supply voltage with 8b-IN, 8b-W, and 20b-OUT. Moreover, the figure-of-merit for BSTCIM is 1.13–33.6 times higher than that of existing CIM macros.