Compiling probabilistic, bio-inspired circuits on a field programmable analog array.

Area of Science:

• Neuromorphic Engineering
• Computational Science
• Analog Computing

Background:

• Digital implementations of dynamical systems are often energy-intensive.
• Probabilistic computing offers potential for enhanced computational efficiency.
• Field Programmable Analog Arrays (FPAs) are adaptable analog circuits.

Purpose of the Study:

• To present a Field Programmable Analog Array (FPAA) as an energy and computational efficiency engine.
• To demonstrate that core computations of dynamical systems can be performed on an FPAA with lower energy costs than digital methods.
• To showcase the FPAA's capability in stochastically simulating complex systems.

Main Methods:

• Implementation of a dynamically controllable stochastic system using voltage-controlled amplifiers and comparator thresholds.
• Computation of Bernoulli random variables and subsequent generation of exponentially and arbitrarily distributed random variables.
• Simulation of the Gillespie algorithm on the FPAA to model a biological system's trajectory.

Main Results:

• The FPAA computes functions at significantly lower energy costs compared to digital implementations.
• The system successfully generates various random variables from a core Bernoulli computation.
• A 127X performance improvement was achieved in simulating a biological system's trajectory using the FPAA compared to software approaches.

Conclusions:

• The presented FPAA serves as a viable energy and computational efficiency engine for dynamical systems.
• Probabilistic computing on FPAAs offers a substantial advantage over traditional digital methods for complex simulations.
• This approach has broad relevance for simulating any dynamical system, with potential applications in various scientific fields.