Design of Switched-Current Based Low-Power PIM Vision System for IoT Applications

Neural networks(NN) is becoming dominant in machine learning field for its excellent performance in classification, recognition and so on. However, the huge computation and memory overhead make it hard to implement NN algorithms on the existing platforms with real-time and energy-efficient performance. In this work, a low-power processing-in-memory (PIM) vision system for accelerate binary weight networks is proposed. This architecture utilizes PIM and features an energy-efficient switched current (SI) neuron, employing a network with binary weight and 9-bit activation. Simulation result shows the design occupies 5.82mm2 in SMIC 180nm CMOS technology, which consumes 1.45mW from 1.8V supplies. Our system outperforms the state-of-the-art designs in terms of power consumption and achieves energy efficiency up to 28.25TOPS/W.

[1]  Anantha Chandrakasan,et al.  Conv-RAM: An energy-efficient SRAM with embedded convolution computation for low-power CNN-based machine learning applications , 2018, 2018 IEEE International Solid - State Circuits Conference - (ISSCC).

[2]  Sujan Kumar Gonugondla,et al.  A 42pJ/decision 3.12TOPS/W robust in-memory machine learning classifier with on-chip training , 2018, 2018 IEEE International Solid - State Circuits Conference - (ISSCC).

[3]  Igor Carron,et al.  XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks , 2016 .

[4]  Franziska Hoffmann,et al.  Design Of Analog Cmos Integrated Circuits , 2016 .

[5]  Yuan Gao,et al.  RedEye: Analog ConvNet Image Sensor Architecture for Continuous Mobile Vision , 2016, 2016 ACM/IEEE 43rd Annual International Symposium on Computer Architecture (ISCA).

[6]  J. Babcock,et al.  Analog characteristics of metal-insulator-metal capacitors using PECVD nitride dielectrics , 2001, IEEE Electron Device Letters.

[7]  John B. Hughes,et al.  Switched currents-a new technique for analog sampled-data signal processing , 1989, IEEE International Symposium on Circuits and Systems,.

[8]  Yoshua Bengio,et al.  BinaryConnect: Training Deep Neural Networks with binary weights during propagations , 2015, NIPS.

[9]  Ninghui Sun,et al.  DianNao: a small-footprint high-throughput accelerator for ubiquitous machine-learning , 2014, ASPLOS.

[10]  Geoffrey E. Hinton,et al.  ImageNet classification with deep convolutional neural networks , 2012, Commun. ACM.

[11]  David J. Allstot,et al.  Switched-current circuit design issues , 1991 .

[12]  Jan Van der Spiegel,et al.  Linear Current-Mode Active Pixel Sensor , 2007, IEEE Journal of Solid-State Circuits.

[13]  Yuhua Cheng,et al.  The influence and modeling of process variation and device mismatch for analog/rf circuit design , 2002, Proceedings of the Fourth IEEE International Caracas Conference on Devices, Circuits and Systems (Cat. No.02TH8611).

[14]  Zhuo Wang,et al.  In-Memory Computation of a Machine-Learning Classifier in a Standard 6T SRAM Array , 2017, IEEE Journal of Solid-State Circuits.

[15]  Leibo Liu,et al.  An Ultra-High Energy-Efficient Reconfigurable Processor for Deep Neural Networks with Binary/Ternary Weights in 28NM CMOS , 2018, 2018 IEEE Symposium on VLSI Circuits.

[16]  Marian Verhelst,et al.  An always-on 3.8μJ/86% CIFAR-10 mixed-signal binary CNN processor with all memory on chip in 28nm CMOS , 2018, 2018 IEEE International Solid - State Circuits Conference - (ISSCC).

[17]  Meng-Fan Chang,et al.  A 65nm 4Kb algorithm-dependent computing-in-memory SRAM unit-macro with 2.3ns and 55.8TOPS/W fully parallel product-sum operation for binary DNN edge processors , 2018, 2018 IEEE International Solid - State Circuits Conference - (ISSCC).

[18]  Shuchang Zhou,et al.  DoReFa-Net: Training Low Bitwidth Convolutional Neural Networks with Low Bitwidth Gradients , 2016, ArXiv.