ABSTRACT
Time-domain (TD) computing has attracted attention for its high computing efficiency and suitability for applications on energy-constrained edge devices. In this paper, we present a time-domain compute-in-memory (TDCIM) macro for binary neural networks (BNNs) realized by standard as well as custom delay cells. Multiply-and-accumulate (MAC) operations, batch normalization (BN) and binarization (Bin) are all processed in the time-domain, avoiding costly digital domain post-processing. In addition, it supports flexible mapping for different kernel sizes, achieving 100% utilization. Starting from a standard cell-based implementation, we propose two custom cells that provide interesting trade-offs between energy efficiency, area and accuracy. The two proposed custom designs can achieve 1.5 and 2.06 POPS/W energy efficiencies at 0.5V and 0.6V with less cell area while maintaining model test accuracy.
- Zhengyu Chen et al. 2019. A Time-Domain Computing Accelerated Image Recognition Processor With Efficient Time Encoding and Non-Linear Logic Operation. IEEE J. Solid-State Circuits 54, 11 (2019), 3226--3237. https://doi.org/ 10.1109/JSSC.2018.2883394Google ScholarCross Ref
- Zhengyu Chen et al. 2020. A Mixed-signal Time-Domain Generative Adversarial Network Accelerator with Efficient Subthreshold Time Multiplier and Mixedsignal On-chip Training for Low Power Edge Devices. In IEEE Symp. on VLSI Circuits. https://doi.org/10.1109/VLSICircuits18222.2020.9162829Google Scholar
- Sai Kiran Cherupally et al. 2022. Improving DNN Hardware Accuracy by In- Memory Computing Noise Injection. IEEE Des. & Test 39, 4 (2022), 71--80. https: //doi.org/10.1109/MDAT.2021.3139047Google ScholarCross Ref
- Yu-Der Chih et al. 2021. An 89TOPS/W and 16.3TOPS/mm2 All-Digital SRAMBased Full-Precision Compute-In Memory Macro in 22nm for Machine-Learning Edge Applications. In IEEE Int. Solid-State Circuits Conf. (ISSCC). 252--253. https: //doi.org/10.1109/ISSCC42613.2021.9365766Google Scholar
- Luke R. Everson et al. 2019. An Energy-Efficient One-Shot Time-Based Neural Network Accelerator Employing Dynamic Threshold Error Correction in 65 nm. IEEE J. Solid-State Circuits 54, 10 (2019), 2777--2785. https://doi.org/10.1109/jssc. 2019.2914361Google ScholarCross Ref
- Xin Fan et al. 2019. Synthesizable Memory Arrays Based on Logic Gates for Subthreshold Operation in IoT. IEEE Trans. Circuits Syst.I, Reg. Papers 66, 3 (2019), 941--954. https://doi.org/10.1109/TCSI.2018.2873026Google ScholarCross Ref
- Florian Freye et al. 2022. Memristive Devices for Time Domain Compute-in- Memory. IEEE J. Explor. Solid-State Comput. Devices and Circuits 8, 2 (2022), 119--127. https://doi.org/10.1109/JXCDC.2022.3217098Google ScholarCross Ref
- Hidehiro Fujiwara et al. 2022. A 5-nm 254-TOPS/W 221-TOPS/mm2 Fully-Digital Computing-in-Memory Macro Supporting Wide-Range Dynamic-Voltage-Frequency Scaling and Simultaneous MAC and Write Operations. In IEEE Int. Solid-State Circuits Conf. (ISSCC). 186--187. https://doi.org/10.1109/ISSCC42614.2022.9731754Google Scholar
- Michael Gansen et al. 2022. Discrete Steps towards Approximate Computing. In Int. Symp. Qual. Electron. Des. (ISQED). https://doi.org/10.1109/ISQED54688.2022.9806215Google Scholar
- Hyunjoon Kim et al. 2021. Colonnade: A Reconfigurable SRAM-Based Digital Bit- Serial Compute-In-Memory Macro for Processing Neural Networks. IEEE J. Solid- State Circuits 56, 7 (2021), 2221--2233. https://doi.org/10.1109/JSSC.2021.3061508Google ScholarCross Ref
- Hyunjoon Kim et al. 2023. A 1--16b Reconfigurable 80Kb 7T SRAM-Based Digital Near-Memory Computing Macro for Processing Neural Networks. IEEE Trans. Circuits Syst.I, Reg. Papers (2023). https://doi.org/10.1109/TCSI.2022.3232648Google Scholar
- Jinseok Kim et al. 2019. Area-Efficient and Variation-Tolerant In-Memory BNN Computing using 6T SRAM Array. In IEEE Symp. on VLSI Circuits. 118--119. https://doi.org/10.23919/VLSIC.2019.8778160Google Scholar
- Gokul Krishnan et al. 2022. Hybrid RRAM/SRAM In-Memory Computing for Robust DNN Acceleration. IEEE Trans. Computer-Aided Design Integr. Circuits Syst. 41, 11 (2022), 4241--4252. https://doi.org/10.1109/TCAD.2022.3197516Google ScholarDigital Library
- Jie Lou et al. 2022. All-Digital Time-Domain Compute-in-Memory Engine for Binary Neural Networks With 1.05POPS/W Energy Efficiency. In IEEE Eur. Solid State Circuits Conf. (ESSCIRC). 149--152. https://doi.org/10.1109/ESSCIRC55480.2022.9911382Google Scholar
- Daisuke Miyashita et al. 2017. A Neuromorphic Chip Optimized for Deep Learning and CMOS Technology With Time-Domain Analog and Digital Mixed-Signal Processing. IEEE J. Solid-State Circuits 52, 10 (2017), 2679--2689. https: //doi.org/10.1109/JSSC.2017.2712626Google ScholarCross Ref
- Jyotishman Saikia et al. 2021. Modeling and Optimization of SRAM-based In-Memory Computing Hardware Design. In IEEE Design Automat. Test Eur. Conf. (DATE). 942--947. https://doi.org/10.23919/DATE51398.2021.9473973Google Scholar
- Aseem Sayal et al. 2020. A 12.08-TOPS/W All-Digital Time-Domain CNN Engine Using Bi-Directional Memory Delay Lines for Energy Efficient Edge Computing. IEEE J. Solid-State Circuits 55, 1 (2020), 60--75. https://doi.org/10.1109/JSSC.2019. 2939888Google ScholarCross Ref
- Karen Simonyan et al. 2015. Very Deep Convolutional Networks for Large- Scale Image Recognition. In Proc. Int. Conf. Learn. Represent. (ICLR). https: //doi.org/arXiv:1409.1556Google Scholar
- Jiahao Song et al. 2021. TD-SRAM: Time-Domain-Based In-Memory Computing Macro for Binary Neural Networks. IEEE Trans. Circuits Syst.I, Reg. Papers 68, 8 (2021), 3377--3387. https://doi.org/10.1109/TCSI.2021.3083275Google ScholarCross Ref
- Tim Stadtmann et al. 2020. From Quantitative Analysis to Synthesis of Efficient Binary Neural Networks. In IEEE Int. Conf. Mach. Learn. Appl. (ICMLA). 93--100. https://doi.org/10.1109/ICMLA51294.2020.00024Google Scholar
- Dewei Wang et al. 2022. DIMC: 2219TOPS/W 2569F2/b Digital In-Memory Computing Macro in 28nm Based on Approximate Arithmetic Hardware. In IEEE Int. Solid-State Circuits Conf. (ISSCC). 266--267. https://doi.org/10.1109/ ISSCC42614.2022.9731659Google Scholar
- Cheng-Xin Xue et al. 2021. A 22nm 4Mb 8b-Precision ReRAM Computing-in- Memory Macro with 11.91 to 195.7 TOPS/W for Tiny AI Edge Devices. In IEEE Int. Solid-State Circuits Conf. (ISSCC). 246--247. https://doi.org/10.1109/ISSCC42613. 2021.9365769Google Scholar
- Shihui Yin et al. 2020. XNOR-SRAM: In-Memory Computing SRAM Macro for Binary/Ternary Deep Neural Networks. IEEE J. Solid-State Circuits 55, 6 (2020), 1733--1743. https://doi.org/10.1109/JSSC.2019.2963616Google Scholar
- Chengshuo Yu et al. 2021. A Zero-Skipping Reconfigurable SRAM In-Memory Computing Macro with Binary-Searching ADC. In IEEE Eur. Solid State Circuits Conf. (ESSCIRC). 131--134. https://doi.org/10.1109/ESSCIRC53450.2021.9567819Google Scholar
- Chengshuo Yu et al. 2022. A 65-nm 8T SRAM Compute-in-Memory Macro With Column ADCs for Processing Neural Networks. IEEE J. Solid-State Circuits 57, 11 (2022), 3466--3476. https://doi.org/10.1109/JSSC.2022.3162602Google ScholarCross Ref
Index Terms
- Scalable Time-Domain Compute-in-Memory BNN Engine with 2.06 POPS/W Energy Efficiency for Edge-AI Devices
Recommendations
BNN-Flip: Enhancing the Fault Tolerance and Security of Compute-in-Memory Enabled Binary Neural Network Accelerators
ASPDAC '24: Proceedings of the 29th Asia and South Pacific Design Automation ConferenceCompute-in-memory based binary neural networks or CiM-BNNs offer high energy/area efficiency for the design of edge deep neural network (DNN) accelerators, with only a mild accuracy reduction. However, for successful deployment, the design of CiM-BNNs ...
Towards High Performance and Accurate BNN Inference on FPGA with Structured Fine-Grained Pruning
ICCAD '22: Proceedings of the 41st IEEE/ACM International Conference on Computer-Aided DesignAs the extreme case of quantization networks, Binary Neural Networks (BNNs) have received tremendous attention due to many hardware-friendly properties in terms of storage and computation. To reach the limit of compact models, we attempt to combine ...
Networking Low-Power Energy Harvesting Devices: Measurements and Algorithms
Recent advances in energy harvesting materials and ultra-low-power communications will soon enable the realization of networks composed of energy harvesting devices. These devices will operate using very low ambient energy, such as energy harvested from ...
Comments