基于忆阻器的神经形态计算--高性能计算-上海期智研究院

人工智能

交叉智能

现代密码学

高性能计算

量子计算与量子人工智能

基于忆阻器的神经形态计算-

蒋力

课题介绍

我们在面向视频理解和图像处理优化以及更通用的神经网络算法方面，提出了一系列计算算法与架构协同优化方法；针对神经网络庞大的参数量提出了极低位宽的混合量化方法而去不需要重新训练模型；针对卷积神经网络，时序神经网络和图神经网络提出了动态剪枝技术以减小运算量与数据搬运。

在神经形态计算芯片架构设计方面，我们设计了一种存储架构数据注册技术，通过优化负载均衡提高了存储阵列可扩展性；利用算法获得的bit稀疏性，提出了利用bit级稀疏性的提高基于忆阻器阵列的神经形态芯片的利用率的两种方法；通过软硬结合的优化手段提高了忆阻器阵列的可靠性。

合作方面，与研究院刘明院士团队合作进行了一次ReRAM芯片MPW流片项目，并取得一定进展。

课题团队成员

课题科研成果

Yunyan Hong,Qiang Xu,Li Jiang.Skimming and Scanning for Untrimmed Video Action Recognition. CISP-BMEI 2021.

Hanchen Guo, Zhehan Lin, Yunfei Gu, Chentao Wu, Li Jiang, Jie Li, Guangtao Xue, Minyi Guo. Lazy-WL: A Wear-aware Load Balanced Data Redistribution Method for Efficient SSD Array Scaling. Cluster 2021.

Feiyang Wu, Zhuoran Song, Jing Ke, Li Jiang, Naifeng Jing, Xiaoyao Liang. IPU: Domain Specific Programmable Parallel Microarchitecture for Image Processing. ISPA 2021.

Fangxin Liu, Wenbo Zhao, Zhezhi He, Zongwu Wang, Yilong Zhao, Tao Yang, Xiaoyao Liang, Naifeng Jing, Li Jiang. SME: ReRAM-based Sparse-Multiplication-Engine to Squeeze-Out Bit Sparsity of Neural Network. ICCD 2021.

Fangxin Liu, Wenbo Zhao, Zhezhi He, Zongwu Wang, Yilong Zhao, Yongbiao Chen, Li Jiang. Bit-Transformer: Transforming Bit-level Sparsity into Higher Preformance in ReRAM-based Accelerator. ICCAD 2021.

Fangxin Liu,Wenbo Zhao,Zhezhi He,Yanzhi Wang,Zongwu Wang,Changzhi Dai, Xiaoyao Liang, Li Jiang. Improving Neural Network Efficiency via Post-training Quantization with Adaptive Floating-Point. ICCV 2021.

Dongyue Li,Tao Yang,Lun Du,Zhezhi He,Li Jiang. AdaptiveGCN: Efficient GCN Through Adaptively Sparsifying Graphs. CIKM 2021.

Tao Yang, Dongyue Li, Yibo Han, Yilong Zhao, Fangxin Liu, Xiaoyao Liang，Zhezhi He,Li Jiang. PIMGCN: A ReRAM-Based PIM Design for Graph Convolutional Network Acceleration. DAC 2021.

Zhuoran Song, Yanan Sun, Lerong Chen, Tianjian Li, Naifeng Jing, Xiaoyao Liang, Li Jiang. ITT-RNA: Imperfection Tolerable Training for RRAM-Crossbar-Based Deep Neural-Network Accelerator. TCAD 2021.

Ziqi Meng,Weikang Qian*,Yilong Zhao,Yanan Sun,Rui Yang,Li Jiang. Digital Offset for RRAM-based Neuromorphic Computing: A Novel Solution to Conquer Cycle-to-cycle Variation. DATE 2021.

Fangxin Liu,Wenbo Zhao, Zongwu Wang,Tao Yang，Li Jiang. IM3A: Boosting Deep Neural Network Efficiency via In-Memory Addressing-Assisted Acceleration. GLSVLSI 2021.

Chen Nie, Jie Lin, Huan Hu, Li Jiang, Xiaoyao Liang, Zhezhi He. Energy-Efficient Hybrid-RAM with Hybrid Bit-Serial based VMM Support. GLSVLSI 2021.

Yilong Zhao,Zhezhi He, Naifeng Jing,Xiaoyao Liang,Li Jiang. Re2PIM: A Reconfigurable ReRAM-Based PIM Design for Variable-Sized Vector-Matrix Multiplication. GLSVLSI 2021.

Xingyi Wang, Yu Li, Yiquan Chen, Shiwen Wang, Yin Du, Cheng He, YuZhong Zhang, Pinan Chen, Xin Li, Wenjun Song, Qiang xu, Li Jiang. On Workload-Aware DRAM Failure Prediction in Large-Scale Data Centers. VTS 2021.

Chaoqun Chu, Yanzhi Wang, Yilong Zhao, Xiaolong Ma, Shaokai Ye, Yunyan Hong, Xiaoyao Liang, Yinhe Han, Li Jiang. PIM-Prune: Fine-Grain DCNN Pruning for Crossbar-Based Process-In-Memory Architecture. DAC 2020.

Zhuoran Song, Jianfei Wang, Tianjian Li, Li Jiang, Jing Ke, Xiaoyao Liang, Naifeng Jing. GPNPU: Enabling Efficient Hardware-Based Direct Convolution with Multi-Precision Support in GPU Tensor Cores. DAC 2020.

Tao Yang, Yunkun Liao, Jianping Shi, Yun Liang, Naifeng Jing, Li Jiang. A Winograd-Based CNN Accelerator with a Fine-Grained Regular Sparsity Pattern. FPL 2020.

Zhuoran Song, Yilong Zhao, Yanan Sun, Xiaoyao Liang, Li Jiang. ESNreram: An Energy-Efficient Sparse Neural Network Based on Resistive Random-Access Memory. GLSVLSI 2020.

分享到

基于忆阻器的神经形态计算-

课题介绍

课题团队成员

赵怿龙

杨涛

刘方鑫

课题科研成果