Researchers at Purdue, Arizona State, and Georgia Tech Present at ICCAD

C-BRIC researchers presented their work at the  2022 International Conference on Computer-Aided Design (ICCAD). The IEEE/ACM-sponsored conference was held from October 30-November 4, 2022, in San Diego, California.

 

Purdue researchers Kang He, Indranil Chakraborty, Cheng Wang, Aradhana Mohan Parvathy, Sarada Krithivasan, Sanchari Sen, Anand Raghunathan, and Kaushik Roy, Arizona State researchers Gokul Krishnan, Zhenyu Wang, Jae-sun Seo, and Yu Cao, and Georgia Tech researchers Zishen Wan and Arijit Raychowdhury presented their C-BRIC work at ICCAD.

 

The Purdue team of Kang He, Indranil Chakraborty, Cheng Wang, and Kaushik Roy's published their work on "Design Space and Memory Technology Co-exploration for In-Memory Computing Based Machine Learning Accelerators" at ICCAD. In this work, they explored IMC macro design space and memory technology. Their work focused on identifying the best design point for each memory type under iso-area budgets. They proposed different modes of ADC operations with distinctive weight mapping schemes to cope with different on-chip area budgets. They showed that under small area budgets, the choice of the best memory technology is determined by its cell area and writing energy. While area budgets are larger, cell area becomes the dominant factor for technology selection. He is currently a PhD student under the direction of Kaushik Roy. Chakraborty is a PhD alumnus of Roy's group and is currently working with Google. Wang is a former Postdoc in Roy's group and an Iowa State University faculty member. Roy is a faculty member at Purdue's Elmore Family Scholl of Electrical and Computer Engineering.

 

Purdue's Aradhana Mohan Parvathy, Sarada Krithivasan, Sanchari Sen, and Anand Raghunathan presented "Seprox: Sequence-based Approximations for Compressing Ultra-Low Precision Deep Neural Networks." In this paper, the group proposes Seprox. Seprox is a new compression scheme that exploits two key observations about ultra-low precision DNNs. Lower precision makes fewer weight values possible, leading to increased incidence of frequently-occurring weight sequences. Also, some weight values occur rarely and can be eliminated by replacing them with similar values. Seprox uses approximation techniques to increase the frequencies of the encoded sequences. Parvathy is a PhD student under the direction of Anand Raghunathan. Krithivasan and Sen are PhD graduates of Raghunathan's group; both currently work at IBM. Raghunathan is a faculty member at Purdue's Elmore Family Scholl of Electrical and Computer Engineering.

 

C-BRIC researchers at Arizona State University Gokul Krishnan, Zhenyu Wang, Jae-sun Seo, and Yu Cao, along with collaborators Chaitali Chakrabarti of Arizona State University and A. Alper Goksoy, Sumit Mandal, and Umit Ogras of the University of Wisconsin, Madison published their work on "Big-Little Chiplets for In-Memory Acceleration of DNNs: A Scalable Heterogeneous Architecture" at the conference. This publication focused on a heterogeneous IMC architecture with big-little chiplets and a hybrid network-on-package (NoP) to optimize utilization, interconnect bandwidth, and energy efficiency. The methodology proposed maps the model onto the big-little architecture such that the early layers in the DNN are mapped to the little chiplets with higher NoP bandwidth, and the subsequent layers are mapped to the big chiplets with lower NoP bandwidth. The method achieves scalable solutions by incorporating a DRAM into each chiplet to support a wide range of DNNs beyond the area limit. Krishnan is an Arizona State PhD graduate from Yu Cao's research group and is currently working at Meta. Wang is a PhD student in Cao's group. Seo and Cao are faculty members at the Arizona School of Electrical, Computer, and Energy Engineering.

 

Georgia Tech C-BRIC researchers Zishen Wan and Arijit Raychowdhury and IBM collaborators Karthik Swaminathan, Pin-Yu Chen, and Nandhini Chandramoorthy presented "Analyzing and Improving Resilience and Robustness of Autonomous Machines." This work focused on exploring the ability of an autonomous system to tolerate or mitigate against errors to ensure its functional safety. They explored the origination of fault sources across the computing stack of autonomous systems, discussed the various fault impacts and fault mitigation techniques of different scales of autonomous systems, and pinpointed the challenges and opportunities for assessing and building next-generation resilient and robust autonomous systems. Wan is a PhD student under the direction of Arijit Raychowdhury, and Raychowdhury is a faculty member at the Georgia Institute of Technology School of Electrical and Computer Engineering.