Upcoming events

CS&E Colloquium - Computing Beyond Moore’s Law: From Heterogeneous Chiplets to 3D Adaptive Computing

The computer science colloquium takes place on Mondays from 11:15 a.m. - 12:15 p.m. This week's speaker, Yu (Kevin) Cao from the Department of Electrical and Computer Engineering at UMN, will be giving a talk titled "Computing Beyond Moore’s Law: From Heterogeneous Chiplets to 3D Adaptive Computing."

Abstract

Enabled by advances in packaging technologies, the paradigm shift from monolithic integration to chiplet-based design is reshaping our computing systems, from data centers to edge AI. Composable chiplets offer scalable, flexible, and energy-efficient solutions for increasingly complex and heterogeneous workloads.

This talk presents our recent research on heterogeneous chiplet design and silicon prototypes. We first introduce HISIM, a simulation framework that integrates analytical models and machine-learning surrogates to evaluate power, temperature, performance, area, and cost (PTPAC) across diverse computing units, network topologies, and memory hierarchies. Building upon HISIM, we demonstrate a library of composable and reusable chiplets, integrated with neural architecture search to enable Pareto-optimal co-design of AI models and hardware architectures.

Beyond architectural optimization, we further introduce an emerging paradigm of 3D adaptive computing, leveraging dense vertical interconnects to enable massively parallel data movement and runtime hardware adaptation. By stacking distributed SRAM banks above reconfigurable processing elements, our 3D test chip in GF 22nm achieves sub-microsecond full-chip reconfiguration, over 1,000× faster than conventional FPGAs. This architecture combines ASIC-like efficiency with FPGA-like flexibility, opening new opportunities for SWaP-constrained applications in robotics, autonomous systems, and edge AI.

Biography

Yu Cao received the Ph.D. degree in electrical engineering from University of California, Berkeley, in 2002. He is now the Louis John Schnell Professor of Electrical and Computer Engineering at the University of Minnesota, Minneapolis, Minnesota. He has published numerous articles and three books on nano-CMOS modeling and physical design. His research interests include neural-inspired computing, hardware-algorithm co-design, and heterogeneous integration. Dr. Cao was a Distinguished Lecturer of the IEEE Circuits and Systems Society. He was a recipient of seven Best Paper Awards and Outstanding Researcher Awards from Intel, IBM, and others. He is a Fellow of the IEEE.
 

CRAY Colloquium - Architecture 2.0: Foundations for AI Agents in Computer System Design

The Cray Distinguished Speaker Series takes place on Mondays from 11:15 a.m. - 12:15 p.m. This week's speaker, Vijay Janapa Reddi from Harvard University, will give a talk titled, "Architecture 2.0: Foundations for AI Agents in Computer System Design."

Abstract

Modern computing systems have reached unprecedented levels of physical and architectural complexity, pushing traditional human-driven design methodologies past their sustainable scaling limits. With the breakdown of Dennard scaling and the rise of post-Moore specialization, architects must navigate an exponentially expanding design space spanning heterogeneous accelerators, 2.5D/3D multi-die packaging, domain-specific instruction sets, and dynamic machine learning software stacks. While architectural complexity compounds exponentially, engineering team bandwidth scales linearly—creating a widening “scissors gap” where design exploration, physical placement, and verification signoff consume unsustainable costs and time. This talk presents Architecture 2.0, a new paradigm for computer system design in which autonomous AI agents serve as first-class, collaborative partners across the entire research, specification, implementation, and signoff lifecycle. While machine learning has traditionally influenced computer architecture through isolated point-heuristics and surrogate optimization, agentic design fundamentally transforms this workflow. AI agents actively synthesize architectural variants, invoke compilers and cycle-accurate simulators inside deterministic execution sandboxes, inspect concrete physical and logical feedback, and autonomously iterate through timing and functional repair loops. Realizing this vision is fundamentally a systems infrastructure challenge, not an AI modeling problem. It demands a shared pre-competitive ecosystem ranging from standardized execution harnesses, deterministic simulation sandboxes, and verifiable feedback interfaces to human-in-the-loop signoff governance that avoids moral crumple zones. Drawing inspiration from the 1979 Mead-Conway VLSI revolution that democratized integrated circuit design via ARPANET and multi-project wafers (MPC79), this talk charts a roadmap for building open datasets, standardized benchmarks, and accessible community shuttles—democratizing physical silicon design and laying the foundations for the next era of intelligent computer architecture.

Biography

Dr. Vijay Janapa Reddi is the Gordon McKay Professor of Electrical and Computer Engineering at Harvard University. His research spans computer architecture, machine learning systems, and autonomous agents, with a particular focus on designing efficient AI systems across the computing spectrum, from large-scale AI infrastructure to physical AI systems. Before joining Harvard, he was an Associate Professor of Electrical and Computer Engineering at the University of Texas at Austin. His work has helped shape emerging areas including ML systems, TinyML, edge AI, and the design and evaluation of efficient AI infrastructure. Beyond academia, Reddi is Vice President and co-founder of MLCommons, the nonprofit organization behind MLPerf and other widely used machine learning benchmarks. He serves on its board of directors, oversees MLCommons Research, and co-led the development of MLPerf, which has grown into an industry standard framework for evaluating machine learning systems across computing platforms ranging from megawatt-scale data centers to microwatt-scale embedded devices. He also serves on the board of the EDGE AI Foundation, where he works to strengthen collaboration among academia, industry, and the broader edge AI community. He has received numerous honors for his research and contributions to computer systems, including the National Academy of Engineering Gilbreth Lectureship, the IEEE TCCA Young Computer Architect Award, the Intel Early Career Faculty Award, and multiple Google Faculty Research Awards. His research has received Best Paper awards at DAC, MICRO, and HPCA, as well as numerous selections for IEEE Micro Top Picks in Computer Architecture. He is a member of both the MICRO and HPCA Halls of Fame. He is also passionate about expanding access to computing and AI education. He created the open-source Machine Learning Systems textbook (mlsysbook.ai) and developed the Tiny Machine Learning, or TinyML, edX series, which has reached more than 100,000 learners worldwide. His educational work focuses on making the principles of modern AI systems accessible through hands-on learning that connects algorithms, software, hardware, and real-world deployment. Reddi received his Ph.D. in Computer Science from Harvard University, his M.S. in Electrical and Computer Engineering from the University of Colorado Boulder, and his B.S. in Computer Engineering from Santa Clara University.

 

Code Freeze 2027 - AgenticSE

At CodeFreeze 2027 — AgenticSE — we will explore how the rapid adoption of AI agents is fundamentally changing the way software is developed, verified, validated and maintained. AgenticSE is a one-day symposium that brings together leading industry practitioners, academic researchers and software engineers to examine the practical applications, breakthroughs and challenges of integrating agents into the software engineering lifecycle.