Growing an Open-Source Hardware Infrastructure
John Hennessy and David Patterson’s Turing lecture proclaimed that a new golden age of computer architecture is upon us—one where specialized hardware accelerators will become ubiquitous and replace general-purpose architectures for the most computationally demanding tasks.
There is however, one small problem: designing hardware is hard, especially if you come from a software background. Hardware design languages operate with the low-level abstractions of gates, wires, and clocks cycles and rely on proprietary, vendor-locked toolchains to accomplish anything. While this may be acceptable for people designing state of the art processor chips, it is clearly untenable in the golden age of computer architecture where rapid iteration of hardware is going to be key.
Our group at the Cornell believes that hardware design should be easy—instead of needing person-years of work to develop a simple accelerator, building one should be a simple weekend job. This vision mirrors the progression of software development—deploying a new website, service, or app is something software engineers consider to be boring, simple tasks that take a few line of python to accomplish. Hardware design should be the same.
