BEGIN:VCALENDAR VERSION:2.0 PRODID:IEEE vTools.Events//EN CALSCALE:GREGORIAN BEGIN:VTIMEZONE TZID:US/Eastern BEGIN:DAYLIGHT DTSTART:20210314T030000 TZOFFSETFROM:-0500 TZOFFSETTO:-0400 RRULE:FREQ=YEARLY;BYDAY=2SU;BYMONTH=3 TZNAME:EDT END:DAYLIGHT BEGIN:STANDARD DTSTART:20211107T010000 TZOFFSETFROM:-0400 TZOFFSETTO:-0500 RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=11 TZNAME:EST END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20211014T142212Z UID:E61D4AA6-A70A-4FDA-847E-049903BCF0D2 DTSTART;TZID=US/Eastern:20211015T130000 DTEND;TZID=US/Eastern:20211015T141500 DESCRIPTION:IEEE Magnetics Society and EMBC Seminar: Using magnetic tunnel junctions to compute like the brain\n\nAbstract: Computers\, originally de signed to do precise numerical processing\, are now widely used to do more \ncognitive tasks. These include categorical challenges like image and voi ce recognition\, as well as robotic tasks like\ndriving a car and making r eal-time decisions based on sensory input. While the human brain does not do precise\nnumerical processing well\, it excels at these other tasks\, l eading researchers to look to the brain for inspiration on\nefficient ways to engineer cognitive computers. Of particular interest are energy and sp ace optimization. Computers\ncan now perform many of these cognitive tasks as well as humans\, and often faster\, but at the cost of much higher\nto tal energy consumption and much greater space. Some improvements are being found from algorithms that are\nmore brainlike\, and some from novel elec tronic devices that emulate features of the brain. However\, the greatest\ nprogress can be found by working simultaneously across the computational stack integrating both.\nMagnetic tunnel junctions have several features t hat make them attractive potential devices for these applications.\nOne fe ature is that they are already integrated into fabrication plants for comp lementary-metal-oxide-semiconductor\n(CMOS) integrated circuits. They can be readily integrated with existing CMOS technology to take advantage of i ts\nmany capabilities. Another feature is that they are multifunctional. W ith only slight changes in fabrication details\,\nthey can be modified to provide non-volatile memory\, truly random thermal fluctuations\, or gigah ertz oscillations.\nMagnetic tunnel junctions can be used as a memory to s tore synaptic weights\, but when the weights change too\nfrequently the en ergy cost of repeatedly writing them becomes inefficient. Reducing the ret ention time of the\nmemory reduces the cost of writing them\, leading to a trade-off between energy efficiency and reliability. The\nseemingly rando m patterns of neural spike trains have inspired a number of computational approaches based on the\nrandom thermal fluctuations of superparamagnetic tunnel junctions. I discuss some of these approaches and the\ndesign choic es we have made in implementing neural networks based on superparamagnetic tunnel junctions.\n\nCo-sponsored by: Virginia Commonwealth University\n\ nSpeaker(s): Mark Stiles\, \n\nRoom: E3229\, Bldg: East Engineering Buildi ng\, 401 W Main Street\, Mechanical and Nuclear Engineering\, E3240\, Mech anical and Nuclear Engineering\, Richmond\, Virginia\, United States\, 232 84 LOCATION:Room: E3229\, Bldg: East Engineering Building\, 401 W Main Street\ , Mechanical and Nuclear Engineering\, E3240\, Mechanical and Nuclear Engi neering\, Richmond\, Virginia\, United States\, 23284 ORGANIZER:rhadimani@vcu.edu SEQUENCE:0 SUMMARY:IEEE Magnetic and EMBS Seminar: Using magnetic tunnel junctions to compute like the brain URL;VALUE=URI:https://events.vtools.ieee.org/m/285611 X-ALT-DESC:Description: <br /><p>IEEE Magnetics Society and EMBC Seminar: U sing magnetic tunnel junctions to compute like the brain</p>\n<p>Abstract: Computers\, originally designed to do precise numerical processing\, are now widely used to do more<br />cognitive tasks. These include categorical challenges like image and voice recognition\, as well as robotic tasks li ke<br />driving a car and making real-time decisions based on sensory inpu t. While the human brain does not do precise<br />numerical processing wel l\, it excels at these other tasks\, leading researchers to look to the br ain for inspiration on<br />efficient ways to engineer cognitive computers . Of particular interest are energy and space optimization. Computers<br / >can now perform many of these cognitive tasks as well as humans\, and oft en faster\, but at the cost of much higher<br />total energy consumption a nd much greater space. Some improvements are being found from algorithms t hat are<br />more brainlike\, and some from novel electronic devices that emulate features of the brain. However\, the greatest<br />progress can be found by working simultaneously across the computational stack integratin g both.<br />Magnetic tunnel junctions have several features that make the m attractive potential devices for these applications.<br />One feature is that they are already integrated into fabrication plants for complementar y-metal-oxide-semiconductor<br />(CMOS) integrated circuits. They can be r eadily integrated with existing CMOS technology to take advantage of its<b r />many capabilities. Another feature is that they are multifunctional. W ith only slight changes in fabrication details\,<br />they can be modified to provide non-volatile memory\, truly random thermal fluctuations\, or g igahertz oscillations.<br />Magnetic tunnel junctions can be used as a mem ory to store synaptic weights\, but when the weights change too<br />frequ ently the energy cost of repeatedly writing them becomes inefficient. Redu cing the retention time of the<br />memory reduces the cost of writing the m\, leading to a trade-off between energy efficiency and reliability. The< br />seemingly random patterns of neural spike trains have inspired a numb er of computational approaches based on the<br />random thermal fluctuatio ns of superparamagnetic tunnel junctions. I discuss some of these approach es and the<br />design choices we have made in implementing neural network s based on superparamagnetic tunnel junctions.</p> END:VEVENT END:VCALENDAR