The demise of Moore’s Law has an important implication: the monetary incentive to invest in repeated upgrades for processors will decrease. The understated consequence of this is that processors will need to be designed with the constraint of longevity (and thus fault tolerance) in mind. Architects must design accelerators which maintain performance and increase longevity while not sacrificing too much area or power even in the presence of a broken component.

Traditionally, accelerators have been implemented in a monolithic manner, meaning, if a single part broke, then the whole accelerator becomes unusable. Certainly this is not fault-tolerant. We propose a propose a hardware design methodology for modular accelerators, \textbf{Non-Newtonian Hardware Design} (NNHW), to improve the fault tolerance of accelerators while not overly sacrificing performance or area. We show concrete results for this system built using Cohort. We additionally propose a set of high level requirements for an actor-based software-hardware description language capable of transforming software to hardware when under pressure (like a non-Newtonian fluid’s transition from soft to hard).