Sairam Sri Vatsavai

Research Staff Member, Computational Science Initiative, Brookhaven National Laboratory. Upton, New York.

I build simulators and learned surrogates for large scientific computing facilities, and I design photonic hardware for AI. Both lines of work ask the same question, which is how to predict the behavior of a system that is too large, too expensive, or too new to measure directly.

Before joining Brookhaven I completed a Ph.D. in Electrical Engineering at the University of Kentucky in 2024, advised by Ishan Thakkar, where I worked on photonic integrated circuit based AI accelerators.

News

Sep 2026

TRASSE submitted to IAAI-27 Latest

Learned simulator surrogates for scheduling-policy search, submitted to the Emerging Applications track.

Aug 2026

COSMOS selected for an LDRD award

Brookhaven will fund the multidimensional photonic computing framework from October 2026 through September 2028. I lead benchmarking and evaluation.

Jun 2026

Talk at PASC 2026 in Bern

Presented the simulator-to-surrogate methodology for performance modeling of distributed computing.

2026

Plenary at ATLAS Software and Computing Week

Grid simulation with CGSim and surrogate modeling with TRASSE.

2026

Poster and talk at ModSim 2026

AI-enabled modeling and simulation for distributed computing, with coauthors across Brookhaven, Pittsburgh, CMU, UMass Amherst, ORNL, and SLAC.

Nov 2025

Best Short Paper at PMBS, SC 2025

Awarded for CGSim, the discrete-event simulator behind the grid modeling work.

Publications

2026

Predicting job turnaround time in large-scale distributed computing environments with graph neural networks

EPJ Research Infrastructures, volume 10, article 14  Paper

2025

CGSim, a SimGrid-based discrete-event simulator for distributed scientific computing Best short paper

PMBS Workshop at SC 2025  Paper

2025

HEANA, a hybrid time-amplitude analog optical accelerator with flexible dataflows for energy-efficient CNN inference

ACM Transactions on Design Automation of Electronic Systems, volume 30, issue 2  Paper

2023

SCONNA, a stochastic computing based optical accelerator for ultra-fast, energy-efficient inference of integer-quantized CNNs

IEEE International Parallel and Distributed Processing Symposium  Paper

2022

Photonic reconfigurable accelerators for efficient inference of CNNs with mixed-sized tensors

IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, volume 41, issue 11  Paper

2020

PROTEUS, rule-based self-adaptation in photonic NoCs for loss-aware co-management of laser power and performance

IEEE/ACM International Symposium on Networks-on-Chip  Paper

Under review and in preparation

2026

TRASSE, learned simulator surrogates for scheduling-policy search at distributed scientific computing facilities

Under review

2026

HOLOS, gradient-based design space exploration for LLM parallelism and hardware co-design

Under review

2026

Analytical modeling of superconducting digital systems for large language model inference and training

Under review, IEEE Transactions on Computers

2026

An agentic extension of LightshowAI with standardized tool interfaces for XANES analysis

In preparation

Research

Modeling and surrogates for distributed computing

Discrete-event simulators reproduce the behavior of production grids, and learned surrogates stand in for them once a search needs thousands of rollouts. The goal is scheduling-policy search at facilities where a single real experiment costs weeks of grid time.

Photonic and beyond-CMOS accelerators

Microring and interferometric meshes can carry many computations at once across wavelength, time, spatial mode, and polarization. My work spans the device models, the mapping from workload to degrees of freedom, and the benchmarking that decides whether any of it beats a GPU.

Projects

REDWOOD

A DOE ASCR project on modeling and optimizing scientific cyberinfrastructure. I build the simulation and surrogate stack, CGSim and TRASSE, and the training pipelines that cover 74 WLCG sites.

COSMOS

A Brookhaven LDRD on simulation-driven co-design of multidimensional photonic computing systems. I am a co-investigator leading benchmarking and evaluation against GPU, TPU, and single-degree-of-freedom photonic baselines.

LightshowAI

An X-ray absorption spectroscopy platform for NSLS-II. I work on federated authentication, streaming beamline data, and the agentic layer that exposes surrogate models and materials databases as tools.

HOLOS

A gradient-descent design space exploration framework that searches parallelism strategy and accelerator geometry together, across CMOS and superconducting digital technologies.

Service and talks

Reviewing

  • IEEE Transactions on Computers
  • IEEE Transactions on Computer-Aided Design
  • SC 2025
  • VLSI Design
  • Great Lakes Symposium on VLSI

Recent talks

  • Simulator-to-surrogate performance modeling for distributed computing PASC 2026, Bern
  • Grid simulation and surrogates for ATLAS ATLAS Software and Computing Week, plenary
  • AI-enabled ModSim for distributed computing ModSim 2026