期刊
IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS
卷 16, 期 6, 页码 1250-1260出版社
IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/TBCAS.2022.3209073
关键词
Compact architecture; neuron integration efficiency; neuromorphic chip; spiking neural network; weight quantization
资金
- NSFC [92064004]
- Sichuan Technology Fund [2018GZDZX0025]
This study proposes a co-designed neuromorphic core based on quantized spiking neural network technology, which offers a new solution to improve neuron integration efficiency and effectively reduces the pressure on core area caused by increasing neuron numbers.
Many efforts have been made to improve the neu-ron integration efficiency on neuromorphic chips, such as us-ing emerging memory devices and shrinking CMOS technology nodes. However, in the fully connected (FC) neuromorphic core, increasing the number of neurons will lead to a square increase in synapse & dendrite costs and a high-slope linear increase in soma costs, resulting in an explosive growth of core hardware costs. We propose a co-designed neuromorphic core (SRCcore) based on the quantized spiking neural network (SNN) technology and compact chip design methodology. The cost of the neuron/synapse module in SRCcore weakly depends on the neuron number, which effectively relieves the growth pressure of the core area caused by increasing the neuron number. In the proposed BICS chip based on SRCcore, although the neuron/synapse module implements 1 similar to 16 times of neurons and 1 similar to 66 times of synapses, it only costs an area of 1.79 x 10(7) F-2, which is 7.9%similar to 38.6% of that in previous works. Based on the weight quantization strategy matched with SRCcore, quantized SNNs achieve 0.05%similar to 2.19% higher accuracy than previous works, thus supporting the design and application of SRCcore. Finally, a cross-modeling application is demonstrated based on the chip. We hope this work will accelerate the develop-ment of cortical-scale neuromorphic systems.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据