Seeking challenging opportunities in VLSI industry in Architecture design/verification and analysis.

Good research track record, hands-on experience in full RTL to GDS flow and AI accelerators.

Publications : Research work 

Recommendation letters are available on request, please write to prashanthhcp@gmail.com

My thesis benchmarks various CNN architectures on traditional CPU and GPU, and characterizes various edge accelerators including DPUCZDX8G (Xilinx zynq SOC), Vision P6 (Cadence DNN IP), edgeTPU and few other accelerators using Resnet based CNNs. I use the recent SOTA solution on ImageNet for edge devices : Once-For-All NAS to search for hardware constrained CNN architectures with a wide range  of compute and memory requirement. Searched architectures are characterized using cycle accurate simulators, analytical performance models or on-device measurement. Using roofline analysis the obtained results are used to benchmark the accelerator architecture with relevance to memory-to-compute capacity, impact on the latency, and energy. Evolutionary search is used to obtain the most efficient network architecture for accuracy, with a given hardware constraint.

I also perform roofline analysis of the most common CNN architectures used today on traditional workstation, edge(mobile) CPU and GPU.

Thesis is available on request.

Furthermore, few additional work :

Algorithmic- computer architecture optimization

• Architecture optimization and profiling of systolic arrays for different workloads.

• Design of accelerator components, Multi-objective accelerator-architecture co-design of Capsule Neural Networks.

Evolutionary circuit, system design

• Standard Cell Library design with arithmetic cells developed using evolutionary algorithms.

• CMOS circuit optimization using heuristic algorithms.

• Design and analysis of log, anti-log circuits.

Approximate Computing

• A Probabilistic Approach to Design Inexact Compressors for Approximate Booth Multipliers.

• A merged hybrid MAC for CNNs.

• Approximation - tunable Floating point multiplier. 

Get in touch at prashanth.c@iiitb.ac.in, prashanthhcp@gmail.com 

 This website is best viewed in Desktop mode.

Note : All information and opinions in this website are my own and do not represent the views of my current/previous employer.