{"603626":{"#nid":"603626","#data":{"type":"news","title":"Mixed-signal Processing Powers Bio-mimetic CMOS Chip to Enable Neural Learning in Autonomous Micro-Robots","body":[{"value":"\u003Cp\u003EIn a recent publication and live demonstration at the International Solid State Circuits Conference (ISSCC), researchers from the Georgia Institute of Technology have used analog processing to squeeze a 3.12Top\/W (average) artificial intelligence processor onto a CMOS (55nm) chip, consuming only 690\u0026mu;W (1.2V), and aimed at self-teaching micro-robots that need to learn their immediate environments. The processor implements \u0026lsquo;reinforcement learning\u0026rsquo; \u0026ndash; a behaviorist psychology-inspired learning algorithm that mimics the way dopamine encourages reward-motivated behavior in human social interactions. The paper is authored by Electrical and Computer Engineering graduate researchers - Anvesha Amravati, Saad Bin Nasir, Sivaram Thangadurai, Insik Yoon and their doctoral advisor Professor Arijit Raychowdhury. The writing team was assisted in a live demonstration by Justin Ting, an undergraduate also in the Department of Electrical and Computer Engineering.\u003C\/p\u003E\r\n\r\n\u003Cp\u003EThe paper, presented on February 12, 2018 at the International Solid-State Circuits Conference (ISSCC), states that the test chip inherits properties of stochastic neural networks and recent advances in Q-learning. Mixed-signal\u0026nbsp;processing was adopted. rather than an all-digital approach, to save area and power. Executing the neural learning based algorithms requires the equivalent of 4 to 8bits (1:16 to 1:256) accuracy, according to the research team, which rules out analogue voltage computation because of the limiting effect of low supply voltage on dynamic range. Instead, analog pulse-widths have been used, thereby enabling large dynamic ranges. As a trade-off, the architectures are slower, but not to the point at which they become unacceptable for the applications in hand\u0026rdquo;.\u003C\/p\u003E\r\n\r\n\u003Cp\u003EAs an example of the mixed signal processing within a time-domain neuron, a time-domain multiply-and-accumulate (MAC) is implemented in a 21-bit counter which multiplies the 6-bit input from a pre-synaptic neuron by the 6-bit weight of the synapse. The counter\u0026rsquo;s input is a pulse whose width is proportional to the input value, and the counter is clocked by a frequency proportional to the learned weighting, with the result that the count is proportional to one multiplied by the other. Using an up\/down counter allows negative values of input to be accommodated.\u003C\/p\u003E\r\n\r\n\u003Cp\u003EThis process looks typically digital up to this point, however the weighing-to-frequency oscillator appears to be based on binary-weighted current sources \u0026ndash; implemented as memory-in-logic to reduce data movement.\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u0026ldquo;The energy to perform a MAC is proportional to the magnitude of the operands and hence the importance of the computation in the neural network, a feature inherent in the brain but missing in digital logic,\u0026rdquo; said the team. The worst-case useable power observed is 1.25pJ\/MAC at 0.8V.\u003C\/p\u003E\r\n\r\n\u003Cp\u003EA micro-robot used to demonstrate the processing algorithm was designed to measure distance using ultra-sonic sensors and to use the 4.5mm\u003Csup\u003E2\u003C\/sup\u003E, 55nm test--chip to control its direction of motion. The measured peak energy efficiency of the developed demonstrator is at 0.8V, with 690pJ\/inference and 1.5nJ\/training cycle.\u003C\/p\u003E\r\n\r\n\u003Cp\u003EAccording to Professor Raychowdhury, \u0026ldquo;This paper presents the first reported integrated circuit which implements reinforcement learning at less than a milli-Watt. This can enable a wide variety of applications in autonomous and bio-mimetic systems.\u0026rdquo;\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cstrong\u003EISSCC paper 7.4\u003C\/strong\u003E\u0026nbsp;\u003Cbr \/\u003E\r\n\u003Cem\u003EA 55nm Time-domain mixed-signal neuromorphic accelerator with stochastic synapses and embedded reinforcement learning for autonomous micro-robot. \u003C\/em\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003ENews Release at Electronics Weekly: www.electronicsweekly.com\/news\/research-news\/isscc-analogue-boost-ai-robot-processor-2018-02\/\u003C\/p\u003E\r\n","summary":null,"format":"limited_html"}],"field_subtitle":"","field_summary":"","field_summary_sentence":[{"value":"\u201cThis paper presents the first reported integrated circuit which implements reinforcement learning at less than a milli-Watt. This can enable a wide variety of applications in autonomous and bio-mimetic systems.\u201d "}],"uid":"27863","created_gmt":"2018-03-12 14:48:42","changed_gmt":"2018-03-20 03:45:51","author":"Christa Ernst","boilerplate_text":"","field_publication":"","field_article_url":"","dateline":{"date":"2018-03-12T00:00:00-04:00","iso_date":"2018-03-12T00:00:00-04:00","tz":"America\/New_York"},"extras":[],"hg_media":{"603624":{"id":"603624","type":"image","title":"Raychowdhury Neural Robotics","body":null,"created":"1520865856","gmt_created":"2018-03-12 14:44:16","changed":"1520865856","gmt_changed":"2018-03-12 14:44:16","alt":"","file":{"fid":"230080","name":"Arijit Neural Robot.png","image_path":"\/sites\/default\/files\/images\/Arijit%20Neural%20Robot.png","image_full_path":"http:\/\/www.tlwarc.hg.gatech.edu\/\/sites\/default\/files\/images\/Arijit%20Neural%20Robot.png","mime":"image\/png","size":544592,"path_740":"http:\/\/www.tlwarc.hg.gatech.edu\/sites\/default\/files\/styles\/740xx_scale\/public\/images\/Arijit%20Neural%20Robot.png?itok=g5oxFh56"}}},"media_ids":["603624"],"groups":[{"id":"213791","name":"3D Systems Packaging Research Center"},{"id":"198081","name":"Georgia Electronic Design Center (GEDC)"},{"id":"197261","name":"Institute for Electronics and Nanotechnology"},{"id":"1271","name":"NanoTECH"},{"id":"213771","name":"The Center for MEMS and Microsystems Technologies"},{"id":"603921","name":"Center for Co-design of Chip Package System"}],"categories":[{"id":"153","name":"Computer Science\/Information Technology and Security"},{"id":"145","name":"Engineering"},{"id":"147","name":"Military Technology"},{"id":"149","name":"Nanotechnology and Nanoscience"},{"id":"152","name":"Robotics"}],"keywords":[{"id":"139771","name":"Arijit Raychowdhury"},{"id":"177356","name":"Center fo Co-design of Chip Package System"},{"id":"177357","name":"C3PS"},{"id":"177119","name":"CAEML"},{"id":"24251","name":"Madhavan Swaminathan"},{"id":"166855","name":"School of Electrical and Computer Engineering"},{"id":"177358","name":"neural learning networks"},{"id":"172977","name":"3D integrated circuits"},{"id":"177359","name":"neuro-architectures"},{"id":"107","name":"Nanotechnology"},{"id":"166968","name":"the Institute for Electronics and Nanotechnology"}],"core_research_areas":[{"id":"39451","name":"Electronics and Nanotechnology"}],"news_room_topics":[],"event_categories":[],"invited_audience":[],"affiliations":[],"classification":[],"areas_of_expertise":[],"news_and_recent_appearances":[],"phone":[],"contact":[{"value":"\u003Cp\u003EChrista Ernst\u003C\/p\u003E\r\n","format":"limited_html"}],"email":["christa.ernst@ien.gatech.edu"],"slides":[],"orientation":[],"userdata":""}}}