Polyarch Research Group

Research Premises and Thrusts

There are two broad premises behind our main research thrusts. First, many architecture families are going to be required to solve future computing system needs. This is largely because of the limitations of technology scaling in providing benefits alone, making architecture innovation and specialization critical. The second is that in order to manage and harness the complexity of these new systems, we will have to rethink and redesign current abstractions between the programmer, software stack, and hardware.

These challenges and our approach manifests differently in different settings. A key determining question is the following: is the application/domain of interest sufficient to justify a complete redesign of the hardware, software interfaces, and programs?

For certain domains (eg. machine learning, big-data processing) the answer would appear to be yes. Here, the question is how to efficiently codesign across all layers of the stack and end up with a design that balances the classic tradeoffs between architecture generality and efficiency. We are currently pursuing the idea of Modular ISAs and automated hardware/software codesign in this space. Read more about the Modular ISA Project here

In most other domains, a redesign of the ISA is impossibly difficult considering legacy software and systems – improving general purpose programmable processors is critical. Here the challenge is to apply the principles of specialization with minimal software disturbance, either with small changes to ISAs or by hiding modifications behind binary translation. Read more about various Transparent Specialization Projects here

Current Students