gkrish19 / SIAM

Scalable In-Memory Acceleration With Mesh: Device, Circuits, Architecture, and Algorithm

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

SIAM

Scalable In-Memory Acceleration With Mesh: Device, Circuits, Architecture, and Algorithm

A comprehensive tool that allows for performance estimation of In-Memory computing (IMC) architectures. The tool is a customized version of NeuroSim (developed by Georgia Tech team led by Prof. Shimeng Yu), combined with a interconnect simulator and a post-mapping accuracy estimator. This tool incorporates device, circuit, architecture, and algorithm properties of in-memory computing. It provides both post-mapping accuracy and hardware performance of IMC architectures. The current version has software and hardware tools separate and will be combined in the coming versions. This version supports Tensorflow (<2.0 version). Software features: Supports quantization, variation-aware training, and hardware-aware training. It incorporates crossbar size, ADC precision, device variations (RRAM only), and algorithm quantization (PACT quantizer). Hardware Features: Incorporates device, circuit, and architecture evaluation and network-on-chip (NoC) evaluation. The current version supports NoC-mesh and will be extended to other topologies.

If you are using this tool, please cite the following work: [1] Krishnan, Gokul, Sumit K. Mandal, Chaitali Chakrabarti, Jae-sun Seo, Umit Y. Ogras, and Yu Cao. "Interconnect-aware area and energy optimization for in-memory acceleration of DNNs." IEEE Design & Test (2020). [2] Mandal, Sumit K., Gokul Krishnan, Chaitali Chakrabarti, Jae-Sun Seo, Yu Cao, and Umit Y. Ogras. "A Latency-Optimized Reconfigurable NoC for In-Memory Acceleration of DNNs." IEEE Journal on Emerging and Selected Topics in Circuits and Systems 10, no. 3 (2020): 362-375.

References: [1] Chen, Pai-Yu, Xiaochen Peng, and Shimeng Yu. "NeuroSim: A circuit-level macro model for benchmarking neuro-inspired architectures in online learning." IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems 37, no. 12 (2018): 3067-3080. [2] Choi, Jungwook, Zhuo Wang, Swagath Venkataramani, Pierce I-Jen Chuang, Vijayalakshmi Srinivasan, and Kailash Gopalakrishnan. "Pact: Parameterized clipping activation for quantized neural networks." arXiv preprint arXiv:1805.06085 (2018).

Acknowledgement: This work was supported in part by C-BRIC, one of the six centers in JUMP, a Semiconductor Research Corporation program sponsored by DARPA.

About

Scalable In-Memory Acceleration With Mesh: Device, Circuits, Architecture, and Algorithm

License:MIT License


Languages

Language:C++ 65.2%Language:Python 34.4%Language:C 0.3%Language:Makefile 0.2%