Computing with Glue, Balls, and Recycled Bits: New Physical Models of Computing

摘要

Real computers live in a physical world, which offers many different ways to compute other than standard CMOS chips. I'll talk about a few different models I've played with, which incorporate geometry and/or thermodynamics into the computational model. 1. Glues: We can build computers out of simple geometric nanoparticles (e.g. DNA) with somewhat selective glues that define how they stick together, or come apart. Our latest result is a way to build a universal replicator, like a photocopier for unknown nanostructures. 2. Balls: We can build computers out of obstacles and rolling balls that just respond to global signals like "roll all balls maximally to the right". The puzzle Tilt embodies this physics, which also have real-world applications in biomedicine. 3. Recycled Bits: We can build computers that recycle bits whenever possible instead of throwing them away a billion times a second. This requires developing a whole new suite of algorithms, but it can lead to several orders of magnitude improved energy efficiency, which ultimately may lead to a similar improvement in speed. 4. In the Limit: What happens when computation and memory are taken to their physical (geometric) limits? I think we should go back to studying algorithms for 1970s-era mesh computers.