Bayesian networks have been thoroughly studied by Judea Pearl, and many algorithms for solving the networks have been proposed [3]. Of these, this paper focuses on the sum-product algorithm, which is a general case of the forward-backward algorithm and many other algorithms developed for artificial intelligence, signal processing, and communications [8].
Carver Mead has championed the field of designing analog VLSI circuits that mimic biological systems. His work is the basis for the circuits presented in this paper and most other previous work. Rather than fighting the non-linearities inherent in transistors, Mead works with them to build precise computational systems. The circuits presented in this paper exploit the exponential characteristics of subthreshold CMOS transistors. Further treatment of these basic circuits is available in [14] and [6].
Much of the background work for this project was drawn from work by Leoliger et al. This Swiss group proposed the idea of performing probability propagation computations in analog VLSI for the purpose of decoding turbo codes and low-density parity-check codes. In their 1998 paper [10], they proposed a set of circuit-level building blocks for creating Trellis modules that decode simple Trellis codes. Their work will be discussed further in section 6.