Elliot Fuller1
Sandia National Laboratories1
Elliot Fuller1
Sandia National Laboratories1
The heterogenous integration of novel materials with existing technology is critical to beyond Moore computing paradigms. Digital computation based upon silicon has relied heavily on a reductionist hierarchy, wherein the system can be reduced to simple rules and elements can designed separately. However, this hierarchy collapses when novel materials are integrated into physics-based compute paradigms to due to the twin difficulties of complexity and scale. Physical compute systems must be co-designed. In 1972, the physicist Philip W. Anderson coined the term, “more is different” to describe emergent phenomena that occur in complex systems where entirely new properties appear, especially where symmetries are broken<sup>1</sup>. Here I will discuss experimental work on building complex physical systems that execute computational kernels, for example outer product updates<sup>2</sup>, and how these systems are “different” from their digital counterparts. Next, I will demonstrate how models can be used to predict the evolution of simple neuromorphic learning systems when operated under certain constraints<sup>3</sup>. These models break down when phase transformation (or symmetry breaking) occurs and the collective behavior of an ensemble of interacting elements is expected to exhibit emergent phenomena. Finally, I will discuss the development of a computing discovery platform to study emergent materials behavior, wherein rapid synthesis, characterization, and integration of new materials into compute arrays enables studying compute outcomes at scale (>1,000 interacting elements).<br/><b>References</b><br/>[1] Anderson <i>Science</i>, <i>177</i>(4047), 393-396 (1972)<br/>[2] Fuller et. al. <i>Science</i>, 364(6440), 570-574 (2019)<br/>[3] Li et. al. <i>Frontiers in Neuroscience</i>, <i>15</i>, 636127 (2021)