Emerging approximate hardware platforms provide operations that, in return for reduced energy consumption, may occasionally produce an incorrect result. We present and evaluate Chisel, a system for reliability-aware optimization of approximate computational kernels that run on such approximate hardware platforms. Given a reliability specification, which identifies a lower bound on the reliability of the result that the kernel computation produces, Chisel automatically determines which kernel operations may execute unreliably and which must execute reliably for the kernel to satisfy its reliability specification.

We evaluate Chisel on five applications from the image processing, scientific computing, and financial analysis domains. The experimental results show that our implemented optimization algorithm enables Chisel to successfully optimize our set of benchmark kernels to obtain energy savings from 8.7% to 19.8% compared to the fully reliable kernel implementations while preserving important reliability guarantees.