CSL Retreat 2025

2025 CSL Retreat

Date: May 7th, 10:45 AM – 6:30 PM
Location: 141 Tata Innovation Center (11 E Loop Rd, New York, NY 10044)

The Computer Systems Laboratory (CSL) at Cornell University consists of a dozen faculty and over 60 PhD students working on hardware and software techniques to improve the cost, performance, programmability, reliability, and energy efficiency of future computing and robotic systems. Our research spans VLSI, RTL, architecture, compilers, and programming languages. This retreat brings our community together to share research, explore innovations in systems, and foster collaboration.

Schedule Overview

Time Session
06:00 AM – 10:30 AM Ithaca members board buses to Cornell Tech
10:45 AM – 11:30 AM Breakfast/Brunch
11:30 AM – 11:45 AM Welcome: Prof. Jaesun Seo
11:45 AM – 12:45 PM Keynote 1: Thijs Roumen (Cornell Tech)
12:45 PM – 1:00 PM Break & Group Picture
01:00 PM – 1:45 PM CSL Group Activity
01:45 PM – 2:15 PM Entrepreneurship and the Runway Post-doc Program: Tomer Joshua (Cornell Tech)
02:15 PM – 2:30 PM Coffee & Snack Break
02:30 PM – 3:30 PM Short Student Talks

  • Allo: A Composable Programming Model for Accelerator Design, Hongzheng Chen
  • A 28nm 20.9-137.2 TOPS/W Output-Stationary SRAM Compute-in-Memory Macro, Xiaofeng Hu
  • TwinSpec: An Asymmetric-Precision Speculative Decoding LLM Accelerator with a Register-Based Vector Quantization Engine, HanGyeol Mun
03:30 PM – 4:30 PM Keynote 2: I-Hsin Chung (IBM Research)
04:30 PM – 5:30 PM Short Student Talks

  • Hermes: Algorithm-System Co-design for Efficient Retrieval-Augmented Generation At-Scale, Michael Shen
  • APT. — Associative Processing Without Breaking the (DRAM) Bank, Cecilio Tamarit
  • Sequence-Level Leakage Risk of Training Data in Large Language Models, Trishita Tiwari
05:30 PM – 6:45 PM Poster Session (Boxed Dinners Available)
06:45 PM – 10:30 PM Return buses to Ithaca (Buses leave at 7)

Keynote Speakers

Thijs Roumen – Cornell Tech

Title: CAD in the Wild

Abstract: In computational fabrication and manufacturing, we have a long tradition of embracing Computer Aided Design (CAD) techniques to design physical objects we end up producing. Inevitably there is a discrepancy between the idealized CAD and real-world conditions of fabrication. In mass manufacturing, we control every machine and environmental parameter to best approximate these idealized conditions. In all other forms of computational fabrication, be it small batch manufacturing or other forms of local distributed manufacturing, we do not have the means to control these conditions. At the Matter of Techlab, we aim to bridge the gap between ideal CAD models and “wild” real-world conditions by building software that embraces this gap. In this talk, I will demonstrate our vision of “CAD in the Wild” at different stages of the fabrication process: before, during, and after fabrication.

Bio: (To Be Announced)

 

I-Hsin Chung – IBM Research

Talk Title: Application-Driven High Performance AI System Design

Abstract: In this talk, we explore the challenges of future system design, focusing on integrating application performance characteristics with emerging technologies to achieve high-performance architecture. We present IBM Vela, a cloud-based HPC system for AI workloads, emphasizing system co-design to approach near bare-metal performance. The talk concludes with research opportunities for future AI system design.

Bio: Dr. I-Hsin Chung, Ph.D. in Computer Science (University of Maryland), is a lead researcher at IBM focusing on performance modeling and HPC/cloud systems. He has contributed to Vela, CORAL, POWER, and Blue Gene and serves as adjunct professor at NYU and Columbia.

Poster Presentations

  • Energy-/Carbon-Aware Evaluation and Optimization of 3D IC Architecture with Digital Compute-in-Memory Designs, Hyung Joon Byun
  • Hybrid Systolic Array Accelerator with Optimized Dataflow for Edge Large Language Model Inference, Chun-Ting Chen
  • Allo: A Composable Programming Model for Accelerator Design, Hongzheng Chen
  • Addressing Carbon Emissions of Computing Across the Stack, Xuesi Chen
  • EcoServe: Designing Carbon-Aware AI Inference Systems, Yueying (Lisa) Li
  • EntoBench: A Benchmark Suite and Evaluation Framework for Insect-Scale Robotics, Derin Ozturk
  • Efficient Scheduling and Data Movement for Accelerated Chip-Multi Processors, Neel Patel
  • NSF Panorama Project: Integrated Rack-Scale Acceleration for Computational Pangenomics, Jan-Niklas Schmelzle, Elton Shih
  • Ground Texture Localization, Aaron Wilhelm
  • Monotone Subsystem Decomposition for Efficient Multi-Objective Robot Design, Andrew Wilhelm
  • Graph Learning at Scale: Characterizing and Optimizing Pre-Propagation GNNs, Zichao Yue

Sponsors & Acknowledgements

Funding for the CSL Retreat was generously provided in part by Bruce Fishbein.
Special thanks to Anais Baez and Anthony Arcangeli for their support in organizing the retreat.