Atthin Chandrashekar

PhD Student · Electrical & Computer Engineering · Purdue University

Advised by Mithuna Thottethodi and T. N. Vijaykumar

Atthin Chandrashekar

About

I'm a second-year PhD student in Electrical and Computer Engineering at Purdue University, advised by Mithuna Thottethodi and T. N. Vijaykumar. I work on computer architecture, with a focus on hardware accelerators and the memory and interconnect systems that feed them.

Most of what I find interesting sits at the boundary between an ML workload and the machine running it - the point where what the model demands stops matching what the hardware was designed to do. Lately that has meant interconnection network analysis, specialized accelerators, and the tooling the field relies on to evaluate architectural ideas before anyone builds them.

Research

Library of Networks (LoN)

An online tool and open database for interconnection network design

Interconnection network design has two decades of published topologies behind it and almost no shared infrastructure for evaluating them. Researchers reimplement the same routing, deadlock-avoidance, and collective-scheduling algorithms just to establish baselines, comparisons across papers are rarely apples-to-apples, and analysis errors in one paper propagate into the next.

Library of Networks pairs a browsable catalog of published on-chip, chiplet, and data center topologies with 27 reusable analysis and synthesis routines, from throughput bounds and deadlock freedom to area and power estimation and cycle-accurate simulation through BookSim 2.0, CNSim, and gem5. Twenty of the twenty-seven have no other public implementation, and the point of collecting them is to make published claims checkable.

My contribution

I designed and built the LoN platform end to end: the web frontend for topology upload, search, comparison, and job submission, and the database backend that stores topologies, jobs, and results. I also designed the Docker worker subsystem that dispatches and executes long-running analysis jobs in the cloud. Because the hosted service and the local deployment run the same container, users can reproduce any published result on their own hardware, use private data that should not leave their machine, or supply their own commercial solver licenses.

Case study

Two interposer topologies from the literature turned out to have identical bisection bandwidth, the field's default proxy for worst-case throughput, contradicting the comparison as published. Maximum concurrent flow, computed in LoN, identified the real difference, and full-system PARSEC simulation confirmed it as an 11% geometric mean speedup.

Teaching

Graduate TA

  • ECE 43700 Computer Architecture Fall 2025, Spring 2026, Fall 2026
  • ECE 27000 Introduction to Digital System Design Summer 2026

Undergraduate TA

  • ECE 43700 Computer Architecture Fall 2024
  • ECE 33700 ASIC Design Spring 2024
  • ECE 20875 Python for Data Science Fall 2023
  • ECE 20001 EE Fundamentals I Spring 2023
  • ENGR 13200 MATLAB Spring 2022

Service

Artifact Evaluation Committee

  • ISCA 2026
  • MICRO 2026

Industry Experience

Caterpillar

Software & Data Engineer Intern Peoria, IL

May 2024 – Aug 2024

Moved CI/CD from Azure DevOps to GitHub Actions, automated integration tests for gen-AI apps, wrote Python/Bash tooling to back up SageMaker notebooks to GitHub, and built a React web/mobile tracker for rental equipment deliveries.

Software & Data Engineer Intern Peoria, IL

May 2023 – Aug 2023

Wrote SQL checks to catch cross-database mismatches across 100+ attributes and shipped a “CAT Inspect” app that generates condition reports in 50+ languages.

Machine Learning Engineer Intern Chennai, India

May 2022 – Aug 2022

Used hidden Markov models in Python to make urban GPS/map-matching less cursed, improving accuracy by 40%+ and reducing GPS response times for navigation in messy city environments.

Education