scieee Open visual document viewer

Editorial: Powering the next-generation IoT applications: new tools and emerging technologies for the development of Neuromorphic System of Systems

Urgese, Gianvito; Ríos-Navarro, Antonio; Linares Barranco, Alejandro; Stewart, Terrence C.; Michmizos, Konstantinos

Published
2023
Publisher
Froontiers Media S.A.
Repository
idus.us.es
Language
en

Abstract

The brain, this 3-pound mass of tissue that can easily be held in one's palm, has an inherent computational complexity that has always inspired efforts to endorse machines with some of its remarkable characteristics. Ironically, the brain computes in its own way, compared to analog or digital computers, despite sharing key concepts with both. It employs analog computation but digital communication, through spikes, both of which improve robustness to noise. This unique combination defines a new computational paradigm that we have just started to explore. The reasons why neuromorphic systems are one of the fastest growing applications are not purely scientific, but mainly technological. For 50 years, the principle guiding computations has been Moore's Law, a macroscopic observation that we will always find ways to engineer faster, smaller and cheaper chips. But there are several reasons why Moore's Law is no longer able to keep up. First, physics: As we downsize transistors close to the atomic scale, it becomes difficult to regulate electron flow. Electrons do not necessarily adhere to Newtonian physics and may pass through the transistor barriers, a phenomenon called quantum tunneling. This makes our computer architectures inefficient. Second, we have long accepted the existence of a trade-off between computing faster and consuming less power, but this has never been a problem until we started approaching the physical limits of fabricating transistors. And the final nail in Moore's Law coffin is put by deep learning. Our computational needs are now orders of magnitude higher than what our systems can deliver. The von-Neumann paradigm is already having more than its fair share of inefficiency—to match its brilliance. And the reason is simple. Computers have been designed with feasibility, not efficiency, at their center. And nowhere are the effects of a bad design more imminent—or the opportunity for an alternative design more compelling – than in emerging technologies, such as edge intelligence, where the computing needs become distilled, asking for real-time solutions to problems constrained by big data. A deep network running on a wearable device will deplete its battery within minutes. The sensors of an autonomous car can easily generate 1 GBps of data. These are examples of the need for real-time computing. The explosive growth of IoT is limited by the efficiency of our computing systems. We are nowhere near ready or prepared for this computational tsunami. There is no better time than this to reconsider the feasibility of alternative solutions. What we need is a computing paradigm that is versatile, robust and power efficient, to handle these seismic shifts in our needs. And what we have now, is enough knowledge on how the brain can achieve these goals. The brain is fault tolerant. It is also extremely power efficient. And it becomes useless if you detach it from its environment: It performs self-learning, one of its most important attributes, by self-organizing based on the input it receives from the environment and other brains. In this topic, we present efforts for advancing non-Von Neumann computations that draw from the brain's functional analogies. Below we walk through the rationale, the challenges and the advantages of redefining algorithms as spiking neural networks, where memory, learning and computing are tightly integrated to advance the implementation of enhanced IoT solutions.

Full text

TYPE Edi o ial PUBLISHED 04 May 2023 DOI 10.3389/ nins.2023.1197918 OPEN ACCESS EDITED AND REVIEWED BY And é an Schaik, Wes e n Sydney Uni e si y, Aus alia *CORRESPONDENCE Gian i o U gese [email p o ec ed] RECEIVED 31 Ma ch 2023 ACCEPTED 07 Ap il 2023 PUBLISHED 04 May 2023 CITATION U gese G, Rios-Na a o A, Lina es-Ba anco A, S ewa TC and Michmizos K (2023) Edi o ial: Powe ing he nex -gene a ion IoT applica ions: new ools and eme ging echnologies o he de elopmen o Neu omo phic Sys em o Sys ems. F on . Neu osci. 17:1197918. doi: 10.3389/ nins.2023.1197918 COPYRIGHT ©2023 U gese, Rios-Na a o, Lina es-Ba anco, S ewa and Michmizos. This is an open-access a icle dis ibu ed unde he e ms o he C ea i e Commons A ibu ion License (CC BY). The use, dis ibu ion o ep oduc ion in o he o ums is pe mi ed, p o ided he o iginal au ho (s) and he copy igh owne (s) a e c edi ed and ha he o iginal publica ion in his jou nal is ci ed, in acco dance wi h accep ed academic p ac ice. No use, dis ibu ion o ep oduc ion is pe mi ed which does no comply wi h hese e ms. Edi o ial: Powe ing he nex -gene a ion IoT applica ions: new ools and eme ging echnologies o he de elopmen o Neu omo phic Sys em o Sys ems Gian i o U gese1*, An onio Rios-Na a o2, Alejand o Lina es-Ba anco2, Te ence C. S ewa 3and Kons an inos Michmizos4 1Poli ecnico di To ino, EDA G oup, To ino, I aly, 2Robo ics and Tech o Compu e s G oup, SCORE Lab, ETSI-EPS, Se illa, Spain, 3Na ional Resea ch Council Canada, O awa, ON, Canada, 4Compu a ional B ain Lab, Depa men o Compu e Science, Ru ge s Uni e si y, Pisca away, NJ, Uni ed S a es KEYWORDS b ain-inspi ed compu a ional p imi i es, neu omo phic enginee ing, neu omo phic IoT applica ions, neu omo phic ools, senso y usion, neu omo phic compu ing, neu omo phic amewo k Edi o ial on he Resea ch Topic Powe ing he nex -gene a ion IoT applica ions: new ools and eme ging echnologies o he de elopmen o Neu omo phic Sys em o Sys ems 1. In oduc ion The b ain, his 3-pound mass o issue ha can easily be held in one’s palm, has an inhe en compu a ional complexi y ha has always inspi ed e o s o endo se machines wi h some o i s ema kable cha ac e is ics. I onically, he b ain compu es in i s own way, compa ed o analog o digi al compu e s, despi e sha ing key concep s wi h bo h. I employs analog compu a ion bu digi al communica ion, h ough spikes, bo h o which imp o e obus ness o noise. This unique combina ion de ines a new compu a ional pa adigm ha we ha e jus s a ed o explo e. The easons why neu omo phic sys ems a e one o he as es g owing applica ions a e no pu ely scien i ic, bu mainly echnological. Fo 50 yea s, he p inciple guiding compu a ions has been Moo e’s Law, a mac oscopic obse a ion ha we will always ind ways o enginee as e , smalle and cheape chips. Bu he e a e se e al easons why Moo e’s Law is no longe able o keep up. Fi s , physics: As we downsize ansis o s close o he a omic scale, i becomes di icul o egula e elec on low. Elec ons do no necessa ily adhe e o New onian physics and may pass h ough he ansis o ba ie s, a phenomenon called quan um unneling. This makes ou compu e a chi ec u es ine icien . Second, we ha e long accep ed he exis ence o a ade-o be ween compu ing as e and consuming less powe , bu his has ne e been a p oblem un il we s a ed app oaching he physical limi s o ab ica ing ansis o s. And he inal nail in Moo e’s Law co in is pu by deep lea ning. Ou F on ie s in Neu oscience 01 on ie sin.o g U gese e al. 10.3389/ nins.2023.1197918 compu a ional needs a e now o de s o magni ude highe han wha ou sys ems can deli e . The on-Neumann pa adigm is al eady ha ing mo e han i s ai sha e o ine iciency— o ma ch i s b illiance. And he eason is simple. Compu e s ha e been designed wi h easibili y, no e iciency, a hei cen e . And nowhe e a e he e ec s o a bad design mo e imminen —o he oppo uni y o an al e na i e design mo e compelling – han in eme ging echnologies, such as edge in elligence, whe e he compu ing needs become dis illed, asking o eal- ime solu ions o p oblems cons ained by big da a. A deep ne wo k unning on a wea able de ice will deple e i s ba e y wi hin minu es. The senso s o an au onomous ca can easily gene a e 1 GBps o da a. These a e examples o he need o eal- ime compu ing. The explosi e g ow h o IoT is limi ed by he e iciency o ou compu ing sys ems. We a e nowhe e nea eady o p epa ed o his compu a ional sunami. The e is no be e ime han his o econside he easibili y o al e na i e solu ions. Wha we need is a compu ing pa adigm ha is e sa ile, obus and powe e icien , o handle hese seismic shi s in ou needs. And wha we ha e now, is enough knowledge on how he b ain can achie e hese goals. The b ain is aul ole an . I is also ex emely powe e icien . And i becomes useless i you de ach i om i s en i onmen : I pe o ms sel -lea ning, one o i s mos impo an a ibu es, by sel -o ganizing based on he inpu i ecei es om he en i onmen and o he b ains. In his opic, we p esen e o s o ad ancing non-Von Neumann compu a ions ha d aw om he b ain’s unc ional analogies. Below we walk h ough he a ionale, he challenges and he ad an ages o ede ining algo i hms as spiking neu al ne wo ks, whe e memo y, lea ning and compu ing a e igh ly in eg a ed o ad ance he implemen a ion o enhanced IoT solu ions. 2. O e iew Wi hin IoT 2.0 and Indus y 4.0 pa adigms, he ansi ion om cloud o edge compu ing is i al o homogeneous and uni e sal da a access h oughou sma connec ed de ices and p oduc li e cycles. This calls o mo e ad anced edge de ices, which o en s uggle wi h powe cons ain s; one o he key playe s in his challenge is he neu omo phic echnology, inspi ed by he mos ad anced and powe -e icien senso da a analy ic sys em— he human b ain. Al hough ini ially in ended o b ain simula ions, he adop ion o he eme ging neu omo phic echnology is mo e and mo e appealing in ields such as IoT edge de ices, Indus y 4.0, biomedical, HPC, and obo ics. This end is con i med by he e o o se e al companies in de eloping neu omo phic a chi ec u es and so wa e ools ha a e opening he way o a new amily o hyb id neu omo phic/digi al IoT de ices ha will gain bene i s om he no el neu o-inspi ed ea u es such as s ochas ici y, low la ency, s uc u al plas ici y, e en -d i en compu a ion, and empo al spa se in o ma ion coding. Neu omo phic compu a ional pa adigms and ha dwa e a chi ec u es a e now ma u e enough o play an impo an ole in IoT applica ions unning on he edge, because o hei abili y o lea n and adap o e e -changing condi ions and asks while espec ing limi ed powe equi emen s. Se e al s a e-o - he-a benchma k applica ions ha e p o ed ha Neu omo phic solu ions, since b ain-inspi ed, p o ide be e scalabili y han adi ional mul i-co e a chi ec u es, and a e especially sui ed o low-powe and adap i e applica ions equi ed o analyze da a in eal- ime. Howe e , he weak s anda diza ion o Neu omo phic componen s, ools, and amewo ks makes i challenging o de ine he enginee ing p ocess o de eloping and o ches a ing hyb idized Neu omo phic/Digi al Sys em o Sys ems deployable in eal-wo ld applica ion scena ios. Ou Resea ch Topic ocus on a ious heo e ical and p ac ical aspec s o di e en Neu omo phic se ups o acili a ing he adop ion o Neu omo phic echnology in o he design o a Sys em o Sys ems p oduc s and algo i hms o IoT applica ions. Pape s in his Resea ch Topic desc ibe he la es ad ances in esea ch on neu omo phic compu a ional pa adigms, encoding algo i hms, amewo ks, oolchains, ools and applica ions which will ac as a e e ence o applica ion de elope s in ol ed in he design o hyb id digi al/neu omo phic sys ems o he IoT domain. Pu a e al. p oposed En o ceSNN a no el design amewo k which enables he design o esilien and ene gy-e icien SNNs conside ing app oxima e DRAMs o embedded sys ems while minimizing hei nega i e impac on he accu acy o he applica ion. The amewo k p o ides a solu ion o esilien and ene gy-e icien SNN in e ence using educed- ol age DRAM o embedded sys ems. Kleijnen e al. epo he capabili y o a ne wo k simula o o p o ide ep esen a i e esul s o he ne wo k load measu ed on he SpiNNake boa d unning a se o benchma k SNNs. Thus, p o iding a powe ul ool o be adop ed in he ea ly phases o he solu ion design. Mülle -Cle e e al. epo on he de ailed implemen a ion o a benchma k o spa io empo al ac ile pa e n ecogni ion a he edge. The au ho s in eg a ed a ull pipeline o implemen ing he B aille le e eading ask using neu omo phic and digi al ools and a chi ec u es. Then, he use case has been analyzed o highligh he s eng hs and weaknesses o he neu omo phic solu ion agains he pu e digi al e sion. Fo no e al. analyzed he signal- o-spike encoding echniques wi h an in o ma ion heo y-based app oach by e alua ing me ics like en opy, mu ual in o ma ion, e iciency, spa si y, and coding e iciency. The analysis has been pe o med on he mos common spike encoding algo i hms using audio and IMU signals as sou ces. Thus, p o iding e e ence indica ions o sys em designe s du ing he enginee ing o he solu ion. Nilsson e al. epo on he cu en landscape o neu omo phic compu ing, ocusing on cha ac e is ics ha pose in eg a ion challenges be ween digi al and neu omo phic echnologies. Based on his analysis, he au ho s p oposed a mic ose ice-based concep ual amewo k o neu omo phic sys ems in eg a ion. Au ho con ibu ions All au ho s lis ed ha e made a subs an ial, di ec , and in ellec ual con ibu ion o he wo k and app o ed i o publica ion. F on ie s in Neu oscience 02 on ie sin.o g U gese e al. 10.3389/ nins.2023.1197918 Funding GU esea ch is pa ially unded by he Eb ains-I aly p ojec CUP B51E22000150006. AR-N and AL-B a e unded by Spanish g an s (wi h suppo om he Eu opean Regional De elopmen Fund) MINDROB (PID2019- 105556GB-C33/AEI/10.13039/501100011033) and SMALL (PCI2019-111841-2/AEI/10.1309/501100011033) p ojec s. Conflic o in e es The au ho s decla e ha he esea ch was conduc ed in he absence o any comme cial o inancial ela ionships ha could be cons ued as a po en ial con lic o in e es . Publishe ’s no e All claims exp essed in his a icle a e solely hose o he au ho s and do no necessa ily ep esen hose o hei a ilia ed o ganiza ions, o hose o he publishe , he edi o s and he e iewe s. Any p oduc ha may be e alua ed in his a icle, o claim ha may be made by i s manu ac u e , is no gua an eed o endo sed by he publishe . F on ie s in Neu oscience 03 on ie sin.o g