REDAC Technology Platform · work in progress
Projects@anabrid
Work running on the REDAC Technology Platform at anabrid. Each project is described here in the open; the measurements and the engineering behind them sit behind a password.
These are problems we are putting on a physical computer rather than stepping through one. The REDAC solves a differential equation by being a system that obeys it, so the question is never only whether the answer is right. It is also what the machine does to the physics on the way, and what has to be done to the machine to get the physics back.
Every project below is work in progress. The public description says what is being investigated and why that is hard on this kind of machine. What is measured, what failed, and what we do about it is behind the password on each page, because most of it is unpublished and some of it belongs to partners. This section is separate from the oscillatory computing work, which is my own research and carries its own password.
-
first results
Battery management systems
A lithium-ion particle exchanges lithium with the electrolyte through its surface alone, so the outside world enters the model as a single boundary term. What it takes to solve that on an analog computer is the question.
-
measured
Gate-based quantum computing on coupled oscillators
Gate-based quantum circuits mapped onto networks of coupled harmonic oscillators, and which circuits survive the mapping.
-
measured
Machine learning on REDAC
Handwritten digits and tetromino shapes classified by an analog substrate, where the physics is fixed and only the readout is fitted.
-
measured
Fluid and aerodynamics
An engine duct written as a partial differential equation and solved as a continuous system rather than stepped on a digital one.
-
measured
The wave equation on strings
A string is the simplest thing that carries a wave, and a direct test of whether the machine keeps the physics it was handed.
-
digital twin measured
The oscillator AI chip
An always-on analog pre-processor built from damped harmonic oscillator pairs, modelled element by element before silicon. A live bench of the architecture runs in the browser behind the password.
-
digital twin measured
Solving PDEs with oscillator networks
A network of damped oscillators, trained on the residual of the viscous Burgers equation and not on data, and what it takes to run it on the REDAC. A live bench of the trained network and of the circuit's cost runs in the browser behind the password.