Neural network hardware performance criteria

Abstract

For many real-world problems where neural networks are used, time constraints make hardware implementation indispensable. This is simply due to the fact that only hardware implementations can fully utilize the benefits coming from the inherent parallelism in neural networks. Our study start with a brief overview of the existing important chips for neural networks from major companies in electronic chip manufactoring shows the enormous diversity among them. Both for this fact and the fact that there is little consensus on the term performance, comparing these chips and choosing one for an application becomes a very difficult task. This work will contribute to that task by generating a new set of criteria which represents the functionality of a neural network chip in a more accurate fashion than the existing criteria.In the main part of the present work, first meta-criteria are proposed as guidelines to come up with a set of good criteria. These meta-criteria demand both generality and practicality of a hardware performance criterium. Then, several commonly used conventional criteria are evaluated according to the proposed meta-criteria. The ineffectiveness of these existing criteria are revealed, from which it can be concluded that there is need for additional hardware performance criteria to make better comparisons possible. We, next, study theoretical aspects of a few new criteria, which are primarily related to the capacity of a neural network. We find that such criteria are often intractable or too weak to say something about performance of hardware, but they can be used as a starting point for more hardware related criteria. Several new hardware performance criteria are proposed, such as a "Reconfigurability Number" that says something about the size and reconfigurability of a chip. Also, we propose a new speed criterium which is normalized to the effective accuracy of a connection: "Effective Connection Primitives Per Second". Finally, experiments with the Intel ETANN chip are reported, where some of the proposed criteria are put into practice. All the mentioned new criteria, together, provide a set which is much more effective than the existing conventional criteria for evaluation of the performance of a neural network chip.

A real-time clustering microchip neural engine

“…In performance comparison of hardware implementations, a common figure of merit is the number of interconnections per second. More refined figures have recently been proposed that include resolution and precision [23]. However, these figures would be reasonably fair criteria for the first type of hardware engineering mentioned above, the general-purpose one.…”

Digital integrated circuit implementations

“…Normalizing the CPS value by the number of weights leads to CPS per weight or CPSPW, and was suggested as a better way to indicate the processing power of a chip (Holler 1991). Precision can also be included in the processing performance by considering a connection primitive per second (CPPS) which is CPS multiplied by bits of precision and by bits for representing the inputs (van Keulan et al 1994). Another reason for taking such speed measurements with a lot of care is that some of the articles report only on a small test chip (and the results reported are extrapolations to a future full-scale chip or to a board of chips and/or neurocomputer), or that only peak values are given.…”

Sistema de hardware reconfigurável para navegação visual de veículos autônomos

“…Existem várias métricas disponíveis para a comparação entre as HNNs ((KEULEN et al, 1994)), porém a maioria dos trabalhos da área utilizam apenas um pequeno grupo para avaliar as implementações. As métricas mais comuns para avaliação de HNNs serão apresentadas de acordo com característica que avaliam.…”

“…Segundo van Keulen existem algumas características necessárias aos critérios de comparação entre HNN ((KEULEN et al, 1994)): o critério deve ser geral, possuir um alto nível conceitual, deve ser independente do problema, deve ter um alto poder de distinção e, o mais importante, os valores utilizados para o cálculo de um critério devem ser possíveis de serem obtidos.…”