Heterogeneous and Intelligent Systems Lab (HISL)

Research Overview

In the post-Moore’s Law era, computing hardware is rapidly diversifying. On the compute side, execution tasks are distributed across CPUs, GPUs, TPUs, and DPUs. On the memory and storage side, characteristics such as latency, bandwidth, capacity, and reliability are becoming increasingly heterogeneous. Simultaneously, upper-layer applications are evolving from traditional programs to LLM-based agents, driving profound shifts in data structures, access patterns, and execution semantics.

HISL explores a fundamental question:

How should system software be retrofitted and redesigned when application semantics become increasingly rich and underlying hardware becomes increasingly heterogeneous?

Centered around this question, Our lab focuses on innovating across operating systems, memory/storage systems, and runtime/compiler layers, specifically rethinking:

• Abstractions: What new primitives should the system provide?
• Interfaces: How should we redefine the boundaries between applications, runtimes, compilers, OS, and hardware?
• Execution Models: What are the optimal execution models for tasks and data across heterogeneous resources?

Team members

T.B.D.