Neuromorphic computing: Difference between revisions

Revision as of 11:51, 21 February 2025

Revision as of 13:21, 7 March 2025

Line 7:

A key aspect of neuromorphic engineering is understanding how the [[Morphology (biology)|morphology]] of individual neurons, circuits, applications, and overall architectures creates desirable computations, affects how [[information]] is represented, influences robustness to damage, incorporates learning and development, adapts to local change (plasticity), and facilitates evolutionary change.

Neuromorphic engineering is an [[Interdisciplinarity|interdisciplinary]] subject that takes inspiration from [[biology]], [[physics]], [[mathematics]], [[computer science]], and [[electronic engineering]]<ref name=":2" /> to design [[Artificial neural network|artificial neural systems]], such as [[Machine vision|vision systems]], head-eye systems, auditory processors, and autonomous robots, whose physical architecture and design principles are based on those of biological nervous systems.<ref>{{Cite journal | doi = 10.1155/2012/705483| title = Qualitative Functional Decomposition Analysis of Evolved Neuromorphic Flight Controllers| journal = Applied Computational Intelligence and Soft Computing| volume = 2012| pages = 1–21| year = 2012| last1 = Boddhu | first1 = S. K. | last2 = Gallagher | first2 = J. C. | doi-access = free}}</ref> One of the first applications for neuromorphic engineering was proposed by [[Carver Mead]]<ref>{{Cite journal |last1=Mead |first1=Carver A. |last2=Mahowald |first2=M. A. |date=1988-01-01 |title=A silicon model of early visual processing %2888%2990024-X |journal=Neural Networks |language=en |volume=1 |issue=1 |pages=91–97 |doi=10.1016/0893-6080(88)90024-X |issn=0893-6080|url=https://resolver.caltech.edu/CaltechAUTHORS:20141223-110732666 }}</ref> in the late 1980s.

==Neurological inspiration==

Line 14:

The biological processes of neurons and their [[synapse]]s are dauntingly complex, and thus very difficult to artificially simulate. A key feature of biological brains is that all of the processing in neurons uses analog [[Cell signalling|chemical signals]]. This makes it hard to replicate brains in computers because the current generation of computers is completely digital. However, the characteristics of these chemical signals can be abstracted into mathematical functions that closely capture the essence of the neuron's operations.{{citation needed|date=February 2025}}

The goal of neuromorphic computing is not to perfectly mimic the brain and all of its functions, but instead to extract what is known of its structure and operations to be used in a practical computing system. No neuromorphic system will claim nor attempt to reproduce every element of neurons and synapses, but all adhere to the idea that computation is highly [[distributed processing|distributed]] throughout a series of small computing elements analogous to a neuron. While this sentiment is standard, researchers chase this goal with different methods.<ref>{{Cite journal | doi = 10.1088/1741-2560/13/5/051001| title = Large-scale neuromorphic computing systems| journal = Journal of Neural Engineering| volume = 13| pages = 1–15| year = 2016| last1 = Furber | first1 = Steve| issue = 5| pmid = 27529195| bibcode = 2016JNEng..13e1001F| doi-access = free}}</ref> Anatomical neural wiring diagrams that are being imaged by electron microscopy<ref>{{cite journal |last1=Devineni |first1=Anita |title=A complete map of the fruit-fly |journal=Nature |date=2 October 2024 |volume=634 |pageissue=358032 |pages=35–36 |doi=10.1038/d41586-024-03029-6|pmid=39358530 }}</ref> and functional neural connection maps that could be potentially obtained via intracellular recording at scale<ref>{{cite journal |last1=Wang |first1=Jun |last2=Jung |first2=Woo-Bin |last3=Gertner |first3=Rona |last4=Park |first4=Hongkun |last5=Ham |first5=Donhee |title=Synaptic connectivity mapping among thousands of neurons via parallelized intracellular recording with a microhole electrode array |journal=Nature Biomedical Engineering |date=2025 |doi=10.1038/s41551-025-01352-5 |pmid=39934437 |url=https://www.nature.com/articles/s41551-025-01352-5}}</ref> can be used to better inspire, if not exactly mimicked, neuromorphic computing systems with more details.

==Implementation==

Line 23:

As early as 2006, researchers at [[Georgia Tech]] published a field programmable neural array.<ref>{{Cite book|last1 = Farquhar|first1 = Ethan|date = May 2006|pages = 4114–4117|last2 = Hasler|first2 = Paul.| title=2006 IEEE International Symposium on Circuits and Systems | chapter=A Field Programmable Neural Array |doi = 10.1109/ISCAS.2006.1693534|isbn = 978-0-7803-9389-9|s2cid = 206966013}}</ref> This chip was the first in a line of increasingly complex arrays of floating gate transistors that allowed programmability of charge on the gates of [[MOSFET]]s to model the channel-ion characteristics of neurons in the brain and was one of the first cases of a silicon programmable array of neurons.

In November 2011, a group of [[MIT]] researchers created a computer chip that mimics the analog, ion-based communication in a synapse between two neurons using 400 transistors and standard [[CMOS]] manufacturing techniques.<ref>{{cite web|title=MIT creates "brain chip"|date=November 15, 2011 |url=http://www.extremetech.com/extreme/105067-mit-creates-brain-chip|access-date=4 December 2012}}</ref><ref name="Neuromorphic silicon paper">{{cite journal|title=Neuromorphic silicon neurons and large-scale neural networks: challenges and opportunities|doi=10.3389/fnins.2011.00108|pmid=21991244|pmc=3181466|volume=5|page=108|journal=Frontiers in Neuroscience|year=2011|last1=Poon|first1=Chi-Sang|last2=Zhou|first2=Kuan|doi-access=free}}</ref>

In June 2012, [[spintronic]] researchers at [[Purdue University]] presented a paper on the design of a neuromorphic chip using [[Spin valve|lateral spin valve]]s and [[memristor]]s. They argue that the architecture works similarly to neurons and can therefore be used to test methods of reproducing the brain's processing. In addition, these chips are significantly more energy-efficient than conventional ones.<ref name="Spin Devices Prop">{{Cite arXiv|title=Proposal For Neuromorphic Hardware Using Spin Devices|eprint=1206.3227|last1=Sharad|first1=Mrigank|last2=Augustine|first2=Charles|last3=Panagopoulos|first3=Georgios|last4=Roy|first4=Kaushik|class=cond-mat.dis-nn|year=2012}}</ref>

Line 43:

The European Union funded a series of projects at the University of Heidelberg, which led to the development of [[BrainScaleS]] (brain-inspired multiscale computation in neuromorphic hybrid systems), a hybrid analog [[neuromorphic]] supercomputer located at Heidelberg University, Germany. It was developed as part of the Human Brain Project neuromorphic computing platform and is the complement to the [[SpiNNaker]] supercomputer (which is based on digital technology). The architecture used in BrainScaleS mimics biological neurons and their connections on a physical level; additionally, since the components are made of silicon, these model neurons operate on average 864 times (24 hours of real time is 100 seconds in the machine simulation) faster than that of their biological counterparts.<ref>{{Cite web|date=2016-03-21|title=Beyond von Neumann, Neuromorphic Computing Steadily Advances|url=https://www.hpcwire.com/2016/03/21/lacking-breakthrough-neuromorphic-computing-steadily-advance/|access-date=2021-10-08|website=HPCwire|language=en-US}}</ref>

In 2019, the European Union funded the project "Neuromorphic quantum computing"<ref>{{Cite journal |title=Neuromrophic Quantum Computing {{!}} Quromorphic Project {{!}} Fact Sheet {{!}} H2020 |url=https://cordis.europa.eu/project/id/828826 |access-date=2024-03-18 |website=CORDIS {{!}} European Commission |language=en |doi=10.3030/828826}}</ref> exploring the use of neuromorphic computing to perform quantum operations. Neuromorphic quantum computing<ref>{{Citation |last1=Pehle |first1=Christian |title=Neuromorphic quantum computing |date=2021-03-30 |arxiv=2005.01533 |last2=Wetterich |first2=Christof|journal=Physical Review E |volume=106 |issue=4 |page=045311 |doi=10.1103/PhysRevE.106.045311 |pmid=36397478 |bibcode=2022PhRvE.106d5311P }}</ref> (abbreviated as 'n.quantum computing') is an [[unconventional computing]] type of computing that uses neuromorphic computing to perform quantum operations.<ref>{{Cite journal |last=Wetterich |first=C. |date=2019-11-01 |title=Quantum computing with classical bits |url=https://www.sciencedirect.com/science/article/pii/S0550321319302627 |journal=Nuclear Physics B |volume=948 |page=114776 |doi=10.1016/j.nuclphysb.2019.114776 |issn=0550-3213|arxiv=1806.05960 |bibcode=2019NuPhB.94814776W }}</ref><ref>{{Citation |last1=Pehle |first1=Christian |title=Emulating quantum computation with artificial neural networks |date=2018-10-24 |arxiv=1810.10335 |last2=Meier |first2=Karlheinz |last3=Oberthaler |first3=Markus |last4=Wetterich |first4=Christof}}</ref> It was suggested that [[quantum algorithm]]s, which are algorithms that run on a realistic model of [[Quantum computing|quantum computation]], can be computed equally efficiently with neuromorphic quantum computing.<ref>{{Cite journal |last1=Carleo |first1=Giuseppe |last2=Troyer |first2=Matthias |date=2017-02-10 |title=Solving the quantum many-body problem with artificial neural networks |url=https://www.science.org/doi/10.1126/science.aag2302 |journal=Science |language=en |volume=355 |issue=6325 |pages=602–606 |doi=10.1126/science.aag2302 |pmid=28183973 |issn=0036-8075|arxiv=1606.02318 |bibcode=2017Sci...355..602C }}</ref><ref>{{Cite journal |last1=Torlai |first1=Giacomo |last2=Mazzola |first2=Guglielmo |last3=Carrasquilla |first3=Juan |last4=Troyer |first4=Matthias |last5=Melko |first5=Roger |last6=Carleo |first6=Giuseppe |date=May 2018 |title=Neural-network quantum state tomography |url=https://www.nature.com/articles/s41567-018-0048-5 |journal=Nature Physics |language=en |volume=14 |issue=5 |pages=447–450 |doi=10.1038/s41567-018-0048-5 |issn=1745-2481|arxiv=1703.05334 |bibcode=2018NatPh..14..447T }}</ref><ref>{{Cite journal |last1=Sharir |first1=Or |last2=Levine |first2=Yoav |last3=Wies |first3=Noam |last4=Carleo |first4=Giuseppe |last5=Shashua |first5=Amnon |date=2020-01-16 |title=Deep Autoregressive Models for the Efficient Variational Simulation of Many-Body Quantum Systems |url=https://link.aps.org/doi/10.1103/PhysRevLett.124.020503 |journal=Physical Review Letters |volume=124 |issue=2 |page=020503 |doi=10.1103/PhysRevLett.124.020503|pmid=32004039 |arxiv=1902.04057 |bibcode=2020PhRvL.124b0503S }}</ref><ref>{{Citation |last1=Broughton |first1=Michael |title=TensorFlow Quantum: A Software Framework for Quantum Machine Learning |date=2021-08-26 |arxiv=2003.02989 |last2=Verdon |first2=Guillaume |last3=McCourt |first3=Trevor |last4=Martinez |first4=Antonio J. |last5=Yoo |first5=Jae Hyeon |last6=Isakov |first6=Sergei V. |last7=Massey |first7=Philip |last8=Halavati |first8=Ramin |last9=Niu |first9=Murphy Yuezhen}}</ref><ref name="Di Ventra">{{Citation |last=Di Ventra |first=Massimiliano |title=MemComputing vs. Quantum Computing: some analogies and major differences |date=2022-03-23 |arxiv=2203.12031}}</ref> Both, traditional quantum computing and neuromorphic quantum computing are physics-based unconventional computing approaches to computations and do not follow the [[von Neumann architecture]]. They both construct a system (a circuit) that represents the physical problem at hand, and then leverage their respective physics properties of the system to seek the "minimum". Neuromorphic quantum computing and quantum computing share similar physical properties during computation.<ref name="Di Ventra"/><ref>{{Cite journal |last1=Wilkinson |first1=Samuel A. |last2=Hartmann |first2=Michael J. |date=2020-06-08 |title=Superconducting quantum many-body circuits for quantum simulation and computing |journal=Applied Physics Letters |volume=116 |issue=23 |doi=10.1063/5.0008202 |issn=0003-6951|arxiv=2003.08838 |bibcode=2020ApPhL.116w0501W }}</ref>

[[Brainchip]] announced in October 2021 that it was taking orders for its Akida AI Processor Development Kits<ref>{{Cite web|url=https://cdn-api.markitdigital.com/apiman-gateway/ASX/asx-research/1.0/file/2924-02438858-2A1332482?access_token=83ff96335c2d45a094df02a206a39ff4|title=Taking Orders of Akida AI Processor Development Kits|date=21 October 2021}}</ref> and in January 2022 that it was taking orders for its Akida AI Processor PCIe boards,<ref>{{cite web | url=https://www.electronics-lab.com/first-mini-pciexpress-board-with-spiking-neural-network-chip/ | title=First mini PCIexpress board with spiking neural network chip | date=January 19, 2022 }}</ref> making it the world's first commercially available neuromorphic processor.