Congratulations to our Ph.D student Shiqi Zhao on his successful graduation!

Congratulations to our Ph.D student Shiqi Zhao on his successful graduation!

As the first Ph.D graduate of CenBRAIN Neurotech, Shiqi Zhao has published 6 papers as first author and co-first author in IEEE Transactions on Biomedical Circuits and Systems, IEEE Transactions on Biomedical Engineeringand other academic journals and conferences during his study.

His Ph.D thesis is entitled“Research on Key Technologies of Artificial Intelligence Chip for Electrophysiological Signal Processing”. The thesis combines hardware-friendly algorithms, processor architecture and circuit design to carry out the key technology research on new electrophysiological signal processor, and the specific work is as follows:

1. To address the trade-off between complexity and accuracy of inference algorithms and the problem of inefficient BP-based supervised training, we proposed energy-efficient inference and training algorithms that can reduce the size of inference models, constructed DFA training models, and realized energy-efficient semi-supervised online learning;

2. To address the shortcomings of traditional physiological signal processor architecture with a single customized model, high energy consumption of always-on mode and support for inference only, a programmable hybrid neural network chip architecture is proposed to realize full-spectrum electrophysiological signal processing and cross-model on-chip few-shot learning while reducing energy consumption.

3. To address the problems of redundant circuit design, dense multiply-accumulate operations and inefficient load mapping, the reconfigurable circuit-level optimization and mapping strategy is proposed to reduce circuit redundancy and simplify the operation path.

4. Based on multiple levels of innovation, we designed and fabricated two chips. The accuracy of cross-signal and cross-patient processing, energy, and area efficiency surpasses all previous related works.