Neuromorphic Circuit Validation Platform for Biomedical Image Computation

Principal Investigator | Darsen Lu, Assistant Professor Department of Electrical Engineering
Co-Principal Investigator |
Nan-Yow, Chen, Associate Research Fellow NAR Labs, NCHC
Wen-Jay Lee, Associate Research Fellow NAR Labs, NCHC
Project Intro
Within the past few years, deep learning have been successfully applied to a wide variety of problems in artificial intelligence. However for very large biomedical image computation problem such as MRI, the training of neural networks will take significant amount of time. The use of GPU is proven to improve the performance of deep learning multi-fold. It essentially overcomes the von-Neumann bottleneck wherein computation is limited by the transfer of data from memory to the CPU. Dedicated neuromorphic IC with non-volatile memory devices and very simple circuitry, with characteristics similar to the biological neurons, enables further massive parallelization. In additional to large-scale bio-medical image processing, IoT device also benefits from deep learning hardware by consuming much less energy. Popular candidates for artificial synapses include resistive RAM and phase-change RAM. The simulation of such system is challenging due to cross-level of abstraction simulation requirement and the need to simulation a large system. In this proposal, in collaboration with NCHC, we would like to build an open simulation platform for neuromorphic circuits to enable design exploration by academic and industry users in Taiwan on both the AI platform and AI application fronts. We would also like to utilize this platform the explore and compare different approaches, including the impact of device and circuit on learning outcome. Finally, we would like to estimate the speed-up from hardware for deep learning on MRI data.
圖片1
Research Method & Techniques
  • Measure characteristics of non-volatile memory (NVM), resistive memory components such as RRAM
  • Build an NVM Model using the measured results
  • Establish a neural network verification platform applied to AI hardware accelerators on NCHC’s cloud application platform-simPlatform
  • Use simulation platform for image recognition of many data sets such as MNIST, CIFAR-10, white blood cells and other biomedical imaging.
Contact info
Tel :06-275-7575 ext.62427
Mail:darsenlu@mail.ncku.edu.tw