Neural Networks and Accelerators

Group Leader: Bing Li

Researchers: Sijie FeiRichard Petri, Zhuorui Zhao, Kangwei XuGuanyu Chi, Ruidi Qiu, Bo LiuTarik Ibrahimpasic, Amro Eldebiky, Wenhao Sun

We are offering reserach topics in the following categories:

  • Hardware accelerators based on digital logic and emerging devices for deep neural networks 
  • Neural network architectures
  • Security of computing hardware and neural networks

Machine learning and neural networks have successfully been applied to solve complex problems such as speech/image processing. To improve computing accuracy, the depth of neural networks has steadily increased significantly, leading to deep neural networks (DNNs). The increasing complexity has put massive demands on computing power and triggered intensive research on hardware acceleration for neuromorphic computing in recent years. To accelerate DNNs, we are exploring solutions based on digital logic, resistive RAM (RRAM), and silicon photonic components. Our research covers novel neural network architectures, efficient mapping of computation operations in DNNs onto hardware accelerators, high-performance and low-power hardware design, and security of computing hardware as well as neural networks.

Digital circuits for accelerating neural networks

In DNNs, the number of required multiply-and-accumulate operations explodes to an extremely huge quantity that resource-constrained platforms cannot afford executing within an acceptable time. To accelerate these operations in DNNs, digital architectures such as systolic array deployed in TPUs have become attractive solutions due to their high degree of concurrent computation and high data reuse rate. In such architectures, weights are first fetched into the processing elements (PEs), and then the input data are streamed into the systolic array, similar to deep pipelining in digital design. To enhance the computation efficiency further, weight pruning has also been developed to compress neural networks without sacrificing accuracy noticeably. This compression takes advantage of the inherent redundancy and flexibility of neural networks, in which a large portion of weights can actually be omitted during the computation. Combining these approaches, we are investigating efficient digital hardware for accelerating neural networks, majorly focusing on circuit architectures, power efficiency, as well as FPGA-based reconfigurable computing.

RRAM-based Neuromorphic Computing

As a promising emerging nonvolatile memory technology, RRAM has been explored intensively for accelerating neuromorphic computing. In such an accelerator, RRAM cells sit at the crossing points of the horizontal wordlines and the vertical bitlines of a crossbar. Multiply operations are implemented by individual RRAM cells according to Ohm’s law, and accumulate operations are implemented by the aggregation of currents according to Kirchhoff’s law. By storing weights of neural networks in memory units and allowing them to participate in computation directly, such in-memory-computing accelerators can overcome the memory bottleneck in the traditional von Neumann architecture. Even more importantly, they exploit the analog nature of the underlying devices to realize the multiply-and-accumulate operations, thus achieving a very high performance and power efficiency. However, due to the analog nature, this computing scheme is very sensitive to hardware uncertainties such as process variations, temperature and noise. To deal with these challenges, we are developing a framework to enhance the robustness and reliability of RRAM-based computing platforms with co-design approaches.

Optical Accelerators for Neural Networks

Optical accelerators for neural networks are implemented with silicon photonic components such as Mach-­Zehnder Interferometers (MZIs) and use light as the computation media.  On the one hand, data propagation inside optical computing units happens at the speed of light. On the other hand, properties of light signals such as phases can be modulated efficiently with a very low power consumption to implement computing functions. Accordingly, computing with light has a large potential to offer high performance and high power efficiency simultaneously. In implementing matrix-vector multiplication operations in neural networks by optical accelerators, the weight matrices need to be decomposed, leading to a rapid increase of the optical components. In addition, hardware uncertainties also affect the computation accuracy severely. To address these problems, we are exploring new architectures of optical accelerators that can offer high performance, robustness, and hardware efficiency simultaneously.

Recent Publications

Richard Petri, Li Zhang, Yiran Chen, Ulf Schlichtmann, Bing Li
PowerPruning: Selecting Weights and Activations for Power-Efficient Neural Network Acceleration
In: ACM/IEEE Design Automation Conference (DAC)
July 2023
PDF (arXiv)

Wenhao Sun, Li Zhang, Huaxi Gu, Bing Li, Ulf Schlichtmann
Class-based Quantization for Neural Networks
In: Design, Automation and Test in Europe (DATE)
April 2023
PDF (arXiv)

Amro Eldebiky, Li Zhang, Georg Böcherer, Bing Li, Ulf Schlichtmann
CorrectNet: Robustness Enhancement of Analog In-Memory Computing for Neural Networks by Error Suppression and Compensation
In: Design, Automation and Test in Europe (DATE)
April 2023
Best Paper Award Candidate
PDF (arXiv)

Wenhao Sun, Li Zhang, Xunzhao Yin, Cheng Zhuo, Huaxi Gu, Bing Li, Ulf Schlichtmann
SteppingNet: A Stepping Neural Network with Incremental Accuracy Enhancement
In: Design, Automation and Test in Europe (DATE)
April 2023
PDF (arXiv)

Bo Liu, Christian Blümm, Stefano Calabrò, Bing Li, Ulf Schlichtmann
Area-Efficient Neural Network CD Equalizer for 4x200Gb/s PAM4 CWDM4 Systems
In: Optical Fiber Communication Conference (OFC)
March 2023

Amro Eldebiky, Georg Böcherer, Maximilian Schaedler, Stefano Calabrò, Bing Li, Ulf Schlichtmann
Implementation of a Robust and Power-Efficient Nonlinear 64-QAM Demapper using In-Memory Computing
In: Optical Fiber Communication Conference (OFC)
March 2023

Tinghuan Chen, Li Zhang, Bei Yu, Bing Li, Ulf Schlichtmann
Machine Learning in Advanced IC Design: A Methodological Survey
IEEE Design & Test

Li Zhang, Bing Li, Xing Huang, Xunzhao Yin, Cheng Zhuo, Masanori Hashimoto, Ulf Schlichtmann
VirtualSync+: Timing Optimization with Virtual Synchronization
IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems (TCAD)
December 2022

Amro Eldebiky, Georg Böcherer, Li Zhang, Bing Li, Maximilian Schaedler, Stefano Calabrò, Ulf Schlichtmann
Power-Efficient and Robust Nonlinear Demapper for 64QAM Using in-Memory Computing
In: European Conference on Optical Communication (ECOC)
September 2022

Di Gao, Qingrong Huang, Li Zhang, Xunzhao Yin, Bing Li, Ulf Schlichtmann, Cheng Zhuo
Bayesian Inference Based Robust Computing on Memristor Crossbar
In: ACM/IEEE Design Automation Conference (DAC)
December 2021

Di Gao, Li Zhang, Xunzhao Yin, Bing Li, Ulf Schlichtmann, Cheng Zhuo
Reliable Memristor-based Neuromorphic Design Using Variation- and Defect-Aware Training
In: IEEE/ACM International Conference on Computer-Aided Design (ICCAD)
November 2021

Li Zhang, Bing Li, Xing Huang, Chen Shen, Shuhang Zhang, Florin Burcea, Helmut Graeb, Tsung-Yi Ho, Hai (Helen) Li, Ulf Schlichtmann
An Efficient Programming Framework for Memristor-based Neuromorphic Computing
In: Design, Automation and Test in Europe (DATE)
February 2021

Jingyao Zhang, Huaxi Gu, Li Zhang, Bing Li, Ulf Schlichtmann
Hardware-Software Codesign of Weight Reshaping and Systolic Array Multiplexing for Efficient CNNs
In: Design, Automation and Test in Europe (DATE)
February 2021

Ying Zhu, Li Zhang, Bing Li, Xunzhao Yin, Cheng Zhuo, Huaixi Gu, Tsung-Yi Ho, Ulf Schlichtmann
Countering Variations and Thermal Effects for Accurate Optical Neural Networks
In: IEEE/ACM International Conference on Computer-Aided Design (ICCAD)
November 2020

Ying Zhu, Li Zhang, Tianchen Wan, Bing Li, Yiyu Shi, Tsung-Yi Ho, Ulf Schlichtmann
Statistical Training for Neuromorphic Computing using Memristor-based Crossbars Considering Process Variations and Noise
In: Design, Automation and Test in Europe (DATE)
March 2020