A neuromorphic approach to adaptive digital circuitry

Walker, M.; Akers, L.A.

doi:10.1109/pccc.1988.10037

Cited by 6 publications

(1 citation statement)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Other approaches for on-chip supervised weight training have been utilized. These approaches include the least-mean-squares algorithm [750], [787], [1025], [1026], weight perturbation [19], [625], [655], [669], [682], [698], [699], [708], [710], [712], [713], [715], [736], [834], [835], [841], [845]- [847], [856], [1078]- [1080], [1098], [1099], [1148], [1304], training specifically for convolutional neural networks [1305], [1306] and others [169], [220], [465], [714], [804], [864], [865], [1029], [1049], [1307]- [1320]. Other on-chip supervised learning mechanisms are built for particular model types, such as Boltzmann machines, restricted Boltzmann machines, or deep belief networks [12], [627], [1135], [1193]<...>…”

Section: A Supervised Learningmentioning

confidence: 99%

A Survey of Neuromorphic Computing and Neural Networks in Hardware

Schuman¹,

Potok²,

Patton³

et al. 2017

Preprint

176

180

View full text Add to dashboard Cite

Neuromorphic computing has come to refer to a variety of brain-inspired computers, devices, and models that contrast the pervasive von Neumann computer architecture. This biologically inspired approach has created highly connected synthetic neurons and synapses that can be used to model neuroscience theories as well as solve challenging machine learning problems. The promise of the technology is to create a brainlike ability to learn and adapt, but the technical challenges are significant, starting with an accurate neuroscience model of how the brain works, to finding materials and engineering breakthroughs to build devices to support these models, to creating a programming framework so the systems can learn, to creating applications with brain-like capabilities. In this work, we provide a comprehensive survey of the research and motivations for neuromorphic computing over its history. We begin with a 35-year review of the motivations and drivers of neuromorphic computing, then look at the major research areas of the field, which we define as neuro-inspired models, algorithms and learning approaches, hardware and devices, supporting systems, and finally applications. We conclude with a broad discussion on the major research topics that need to be addressed in the coming years to see the promise of neuromorphic computing fulfilled. The goals of this work are to provide an exhaustive review of the research conducted in neuromorphic computing since the inception of the term, and to motivate further work by illuminating gaps in the field where new research is needed.

show abstract

Section: A Supervised Learningmentioning

confidence: 99%

A Survey of Neuromorphic Computing and Neural Networks in Hardware

Schuman¹,

Potok²,

Patton³

et al. 2017

Preprint

176

180

View full text Add to dashboard Cite

show abstract

A Synthetic Neural Chip

Grondin

Akers

Walker

et al.

Twenty-Second Asilomar Conference on Signals, Systems and Computers

View full text Add to dashboard Cite

A compact synthetic neural cell that is compatible with CMOS technology is described. Some issues related to its applications are discussed. Particular emphasis is placed on training algorithms and connectivity. Neural Networks and ULSIMicroprocessor chips with over a million transistors and memory chips with over 16 million transistors have already been reported. Simple extrapolations of fabrication technology should allow for a factor of about four in increased density and as wafer scale integration is implemented increases of perhaps two orders of magnitude in die size are also possible. The resulting combination of density and die size extrapolates to well over a billion transistors on one single die. A number of perplexing questions can be raised about such a chip. How do we design it in a reasonable period of time? How can we keep the yield from becoming zero without imposing a requirement that the failure rate per transistor approach one in a trillion? How do we test this chip?The last question may be the most important. What do we want to do with a billion transistors?Biological neural networks are an existence proof that very complex systems, structured and styled in a fashion very different from conventional computers, can perform a variety of useful functions. They offer potential answers to many of the above questions, particularly the questions concerning yield and applications. The features of these biological networks include large numbers (10 to 10 ) of elements that function in a concurrent, asynchronous fashion. These elements have large fan-ins and fan-outs (perhaps as high as lo5). This fact confuses some who mistake this for complete global connectivity, a feature which biological systems in fact do not exhibit. Even with the fan-ins and fan-outs quoted above, in a human each neuron is connected to somewhat less than a millionth of all the neurons in the system. A second feature which is also sometimes ignored is that much of the connectivity is purposeful and that the system is highly structured into recognizable susbsystems that perform recognizable functions e.g. the olfactory bulb. 12 85287-6206While we simply do not know how memory is implemented in a physiological fashion, these systems seem to utilize a non-local or distributed data representation and interaction. The hypothesis most commonly used in neural network analysis and modeling is that learning is associated with some plasticity in the synaptic connections between two neurons. Lastly, a neuron is much more complex than a transistor.Several key features of the above systems can be implemented in silicon. We obviously have large numbers of elements. We can and in fact do build circuits with purposeful connections and distributed data representations.(The use of error correcting codes is an example of a purposeful, distributed data representation.) Plasticity in interconnections is relatively easily performed in a volatile fashion and approaches for nonvolatile but plastic interconnect do exist. We are developing chips that use t...

show abstract