Philip Canoza

EECS Department, University of California, Berkeley

Technical Report No. UCB/EECS-2023-3

January 13, 2023

http://www2.eecs.berkeley.edu/Pubs/TechRpts/2023/EECS-2023-3.pdf

Ising machines have emerged as a new paradigm in unconventional computing that can solve NP-Hard problems. This report looks at both the hardware and algorithm design of these Ising machines. First we will take a look at the implementations of two different hardware accelerators. The first accelerator is a digitally-synthesized, Field Programmable Gate Array (FPGA) that performs synchronous Gibbs sampling of a Restricted Boltzmann Machine (RBM). The second accelerator is an asynchronous neural network that uses a mixed-analog signal neuron to drive stochastic local updates. On the algorithmic side, we explore mappings of the Travelling Salesman Problem. In particular, we introduce a Quadratic Unconstrained Binary Optimization (QUBO) mapping that uses the classical k-opt algorithm to perform local search.

Advisors: Sayeef Salahuddin

---

BibTeX citation:

@mastersthesis{Canoza:EECS-2023-3,
    Author= {Canoza, Philip},
    Title= {Hardware Accelerators and Optimization Algorithms for Unconventional Computing},
    School= {EECS Department, University of California, Berkeley},
    Year= {2023},
    Month= {Jan},
    Url= {http://www2.eecs.berkeley.edu/Pubs/TechRpts/2023/EECS-2023-3.html},
    Number= {UCB/EECS-2023-3},
    Abstract= {Ising machines have emerged as a new paradigm in unconventional computing that can solve NP-Hard problems.  This report looks at both the hardware and algorithm design of these Ising machines.  First we will take a look at the implementations of two different hardware accelerators. The first accelerator is a digitally-synthesized, Field Programmable Gate Array (FPGA) that performs synchronous Gibbs sampling of a Restricted Boltzmann Machine (RBM).  The second accelerator is an asynchronous neural network that uses a mixed-analog signal neuron to drive stochastic local updates.  On the algorithmic side, we explore mappings of the Travelling Salesman Problem. In particular, we introduce a Quadratic Unconstrained Binary Optimization (QUBO) mapping that uses the classical k-opt algorithm to perform local search.},
}

EndNote citation:

%0 Thesis
%A Canoza, Philip 
%T Hardware Accelerators and Optimization Algorithms for Unconventional Computing
%I EECS Department, University of California, Berkeley
%D 2023
%8 January 13
%@ UCB/EECS-2023-3
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/2023/EECS-2023-3.html
%F Canoza:EECS-2023-3