Assembling Integrated Electronics

Modern high-performance computing (HPC) systems consist of static architectures built from monolithic components. Miniaturization driven by lithographic technology has pushed Moore’s Law to its limit after more than half a century, to the point that new chips require multi-billion dollar investments and supercomputer systems are built on a decades-long planning horizon. At the same time, typical HPC workloads like physical simulation have inherent geometry which is not reflected in the compute architecture, leading to a broad range of issues from cache concurrency to programming difficulty. Beyond integrated circuits, adjacent problems exist in electronics generally; printed circuit board assemblies (PCBAs) are similarly static, and the production and recycling of these products is environmentally unsustainable and requires extensive infrastructure. The solution is to modularize electronics and autonomously assemble 3-dimensional computing structures from asynchronous, reusable elements. Of course, this concept brings with it a host of new questions: how are the devices programmed, how is communication bandwidth conserved, how do the elements physically interact, and how are the structures fabricated and assembled? This thesis provides insight on module design and assembly automation for 3-dimensional electronics through two distinct prototype iterations. Evaluation of these systems revealed the mechanical limitations of commercial connectors, so an alternative method called digital materials is described which merges electrical interconnect and physical substrate. This method discretizes substrates into the fundamental elements that make up interconnect systems: conductive and insulating parts which are properly arranged to route signals to asynchronous processing nodes. Along the way, a novel method for constraining motion in these discrete assembly systems using modular superelastic flexures is introduced, characterized, and used to rapidly fabricate several machines.S.M