{"id":3485,"date":"2019-05-09T15:15:52","date_gmt":"2019-05-09T06:15:52","guid":{"rendered":"http:\/\/163.180.4.222\/lab\/?p=3485"},"modified":"2019-05-09T15:15:52","modified_gmt":"2019-05-09T06:15:52","slug":"a-role-for-optics-in-ai-hardware","status":"publish","type":"post","link":"https:\/\/biochemistry.khu.ac.kr\/lab\/?p=3485","title":{"rendered":"A role for optics in AI hardware"},"content":{"rendered":"<p>&nbsp;<\/p>\n<p>&nbsp;<\/p>\n<h5>Experiments show how an all-optical version of an artificial neural network \u2014 a type of artificial-intelligence system \u2014 could potentially deliver better energy efficiency can conventional computing approaches.<\/h5>\n<p>&nbsp;<\/p>\n<div class=\"article__aside align-right hide-print\">\n<div class=\"pdf__download shrink--aside\"><span style=\"color: #82868b; font-size: 1rem;\">Optical fibres transmit data across the world in the form of light and are the backbone of modern telecommunications<\/span><sup style=\"color: #82868b;\"><a href=\"https:\/\/www.nature.com\/articles\/d41586-019-01406-0?utm_source=feedburner&amp;utm_medium=feed&amp;utm_campaign=Feed%3A+nature%2Frss%2Fcurrent+%28Nature+-+Issue%29#ref-CR1\">1<\/a><\/sup><span style=\"color: #82868b; font-size: 1rem;\">. However, when such data need to be analysed, they get converted from light into electrons and are then processed using electronics. There was a time when optics was considered as the basis for a potential computing technology<\/span><sup style=\"color: #82868b;\"><a href=\"https:\/\/www.nature.com\/articles\/d41586-019-01406-0?utm_source=feedburner&amp;utm_medium=feed&amp;utm_campaign=Feed%3A+nature%2Frss%2Fcurrent+%28Nature+-+Issue%29#ref-CR2\">2<\/a><\/sup><span style=\"color: #82868b; font-size: 1rem;\">, but it became difficult for optical computing to compete with the rapid improvements made by its electronic counterpart. In the past few years, however, concern has been growing about the energy costs of computation. Therefore, optics is receiving attention again, both as a way to decrease energy requirements<\/span><sup style=\"color: #82868b;\"><a href=\"https:\/\/www.nature.com\/articles\/d41586-019-01406-0?utm_source=feedburner&amp;utm_medium=feed&amp;utm_campaign=Feed%3A+nature%2Frss%2Fcurrent+%28Nature+-+Issue%29#ref-CR3\">3<\/a><\/sup><span style=\"color: #82868b; font-size: 1rem;\">, and as a special-purpose hardware for accelerating artificial-intelligence algorithms such as deep neural networks (DNNs).\u00a0<\/span><a style=\"background-color: #ffffff; font-size: 1rem;\" href=\"https:\/\/www.nature.com\/articles\/s41586-019-1157-8\" data-track=\"click\" data-label=\"https:\/\/www.nature.com\/articles\/s41586-019-1157-8\" data-track-category=\"body text link\">Writing in\u00a0<i>Nature<\/i><\/a><span style=\"color: #82868b; font-size: 1rem;\">, Feldmann\u00a0<\/span><i style=\"color: #82868b; font-size: 1rem;\">et al.<\/i><sup style=\"color: #82868b;\"><a href=\"https:\/\/www.nature.com\/articles\/d41586-019-01406-0?utm_source=feedburner&amp;utm_medium=feed&amp;utm_campaign=Feed%3A+nature%2Frss%2Fcurrent+%28Nature+-+Issue%29#ref-CR4\">4<\/a><\/sup><span style=\"color: #82868b; font-size: 1rem;\"><span style=\"color: #82868b; font-size: 1rem;\">\u00a0report an intriguing advance towards all-optical implementations of such networks.<\/span><\/span>&nbsp;<\/p>\n<\/div>\n<\/div>\n<div class=\"align-left\">\n<div class=\"article__body serif cleared\">\n<aside class=\"recommended pull pull--left sans-serif\" data-label=\"Related\"><a href=\"https:\/\/www.nature.com\/articles\/s41586-019-1157-8\" data-track=\"click\" data-track-label=\"recommended article\"><img decoding=\"async\" class=\"recommended__image\" src=\"https:\/\/media.nature.com\/w400\/magazine-assets\/d41586-019-01406-0\/d41586-019-01406-0_16690730.jpg\" \/><\/a><\/p>\n<p class=\"recommended__title serif\">Read the paper: All-optical spiking neurosynaptic networks with self-learning capabilities<\/p>\n<\/aside>\n<p>&nbsp;<\/p>\n<p>&nbsp;<\/p>\n<p>A DNN comprises many layers of artificial neurons and artificial synapses, which are connections between the neurons. The strengths of these connections are called weights and can be either positive, indicating neuronal excitation, or negative, implying inhibition. A DNN learns to perform tasks such as image recognition by varying its synaptic weights in a way that minimizes the difference between its actual output and the desired output.<\/p>\n<p>Central processing units and other digital-based hardware accelerators<sup><a href=\"https:\/\/www.nature.com\/articles\/d41586-019-01406-0?utm_source=feedburner&amp;utm_medium=feed&amp;utm_campaign=Feed%3A+nature%2Frss%2Fcurrent+%28Nature+-+Issue%29#ref-CR5\">5<\/a><\/sup>are typically used for DNN computations. A DNN can be trained using a known set of data, whereas an already trained DNN can be applied to unknown data in a task called inference. In either case, although the amount of computation is vast, the variety of operations is modest, because \u2018multiply\u2013accumulate\u2019 operations dominate across the many synaptic weights and neuronal excitations.<\/p>\n<p>DNNs are known to still work well when computational precision is low<sup><a href=\"https:\/\/www.nature.com\/articles\/d41586-019-01406-0?utm_source=feedburner&amp;utm_medium=feed&amp;utm_campaign=Feed%3A+nature%2Frss%2Fcurrent+%28Nature+-+Issue%29#ref-CR5\">5<\/a><\/sup>. As a result, these networks represent an intriguing opportunity for unconventional computing techniques. For example, researchers are exploring DNN accelerators that are based on emerging non-volatile memory devices<sup><a href=\"https:\/\/www.nature.com\/articles\/d41586-019-01406-0?utm_source=feedburner&amp;utm_medium=feed&amp;utm_campaign=Feed%3A+nature%2Frss%2Fcurrent+%28Nature+-+Issue%29#ref-CR6\">6<\/a><\/sup><sup>,<\/sup><sup><a href=\"https:\/\/www.nature.com\/articles\/d41586-019-01406-0?utm_source=feedburner&amp;utm_medium=feed&amp;utm_campaign=Feed%3A+nature%2Frss%2Fcurrent+%28Nature+-+Issue%29#ref-CR7\">7<\/a><\/sup>. Such devices retain information even when their power source is switched off, and can offer improved speed and energy efficiency for DNNs through analog electronic computation.<\/p>\n<p>Why not, therefore, also consider optics? Structures that direct light \u2014 whether they be an optical fibre for use in telecommunications or a waveguide patterned onto a photonic chip \u2014 can be packed with vast amounts of data. Inside such a waveguide, many wavelengths of light can propagate together, using a technique known as wavelength division multiplexing. Each wavelength can then be modulated (altered in such a way that it can carry information) at a rate that is limited by the available bandwidths associated with electronic-to-optical modulation and optical-to-electronic detection.<\/p>\n<p>Structures called resonators enable individual wavelengths to be added to or removed from the waveguide, like wagons on a freight train. For example, micrometre-scale, ring-shaped (micro-ring) resonators can implement arrays of synaptic weights<sup><a href=\"https:\/\/www.nature.com\/articles\/d41586-019-01406-0?utm_source=feedburner&amp;utm_medium=feed&amp;utm_campaign=Feed%3A+nature%2Frss%2Fcurrent+%28Nature+-+Issue%29#ref-CR8\">8<\/a><\/sup>. Such resonators can be modulated thermally<sup><a href=\"https:\/\/www.nature.com\/articles\/d41586-019-01406-0?utm_source=feedburner&amp;utm_medium=feed&amp;utm_campaign=Feed%3A+nature%2Frss%2Fcurrent+%28Nature+-+Issue%29#ref-CR9\">9<\/a><\/sup>, electro-optically<sup><a href=\"https:\/\/www.nature.com\/articles\/d41586-019-01406-0?utm_source=feedburner&amp;utm_medium=feed&amp;utm_campaign=Feed%3A+nature%2Frss%2Fcurrent+%28Nature+-+Issue%29#ref-CR10\">10<\/a><\/sup><sup>,<\/sup><sup><a href=\"https:\/\/www.nature.com\/articles\/d41586-019-01406-0?utm_source=feedburner&amp;utm_medium=feed&amp;utm_campaign=Feed%3A+nature%2Frss%2Fcurrent+%28Nature+-+Issue%29#ref-CR11\">11<\/a><\/sup>\u00a0or, as in Feldmann and colleagues\u2019 work, through phase-change materials<sup><a href=\"https:\/\/www.nature.com\/articles\/d41586-019-01406-0?utm_source=feedburner&amp;utm_medium=feed&amp;utm_campaign=Feed%3A+nature%2Frss%2Fcurrent+%28Nature+-+Issue%29#ref-CR12\">12<\/a><\/sup>. These materials can switch between an amorphous phase and a crystalline phase, which differ greatly in their ability to absorb light. Under ideal conditions, the resulting multiply\u2013accumulate operations would require only a small amount of power.<\/p>\n<p>Feldmann\u00a0<i>et al.<\/i>\u00a0present an all-optical neural network on a millimetre-scale photonic chip, in which there are no optical-to-electronic conversions within the network. Inputted data are electronically modulated onto different wavelengths for injection into the network, but after that has been performed, all the data stay on the chip. Both weight modulation and neuron integration are achieved using integrated phase-change materials; these are located on two types of micro-ring resonator, which have a synaptic or neuronal function.<\/p>\n<p>Unmodulated light that is injected at the various operating wavelengths picks up the neuronal excitations that have accumulated in the phase-change material, and then passes them to the next layer of the network. Even without on-chip optical gain (a process in which a medium transfers energy to the light that is transmitted through it), this set-up could potentially be scaled up to larger networks. The authors demonstrate, on a small scale, both supervised and unsupervised learning \u2014 that is, training is achieved using labelled data, which is how DNNs learn, and using data without such labels, which is how humans tend to learn.<\/p>\n<p>Because the weights are implemented by light absorption, negative weights require a large bias signal, which must not activate the phase-change material. An alternative approach<sup><a href=\"https:\/\/www.nature.com\/articles\/d41586-019-01406-0?utm_source=feedburner&amp;utm_medium=feed&amp;utm_campaign=Feed%3A+nature%2Frss%2Fcurrent+%28Nature+-+Issue%29#ref-CR13\">13<\/a><\/sup>\u00a0that can readily offer negative weights uses devices called Mach\u2013Zehnder interferometers. In these devices, a single waveguide is split into two arms and then recombined; this causes the amount of transmitted light to depend on the difference in optical phase between the two paths. However, it might be challenging to combine this approach with wavelength division multiplexing, because the arms of each interferometer would need to introduce the appropriate phase difference for each wavelength.<\/p>\n<p>Photonic DNNs still present substantial challenges. Their total power usage can be low in ideal situations, but thermo-optic power is frequently required to adjust and maintain the differences in optical phase in the arms of each Mach\u2013Zehnder interferometer. Moreover, the total optical power that is injected into a system containing phase-change materials must be calibrated carefully, so that the materials respond to incoming signals exactly as intended. Although phase-change materials can also be used to adjust Mach\u2013Zehnder phases, unavoidable cross-coupling between how strongly the materials absorb light and how much they slow it down poses a considerable complication.<\/p>\n<p>Phase-change materials seem to be well suited for the non-volatile long-term storage of synaptic weights that are based on micro-ring resonators needing only infrequent adjustment. However, when used in the role of neuron, the speed of crystallization of such materials will limit the maximum rate at which neurons can be excited. Furthermore, the need to melt the materials to induce a full neuronal reset after every potential excitation event will rapidly consume the large, but finite, switching endurance of the materials.<\/p>\n<p>Conventional DNNs have grown large and now typically involve many thousands of neurons and millions of synapses. But photonic networks require waveguides that are spaced far from each other to prevent them from coupling, and that avoid sharp bends to prevent light from leaving the waveguide. Because crossing two waveguides introduces the risk of injecting undesired power into the wrong path, the 2D nature of a photonic chip presents a substantial design constraint.<\/p>\n<p>Despite the long distances and large areas that are required for the implementation of photonic networks, fabrication of the key parts of each optical structure requires precision. This is because the waveguides and coupling regions \u2014 for instance, at the entrance and exit of each micro-ring resonator \u2014 must have the exact dimensions needed to obtain their desired performance. There are also limits to how small micro-ring resonators can be made. Finally, the relatively weak optical effects offered by modulation techniques require long interaction regions to enable their limited impact on passing light to build to a noticeable level.<\/p>\n<p>Advances such as those made in Feldmann and colleagues\u2019 study and by others<sup><a href=\"https:\/\/www.nature.com\/articles\/d41586-019-01406-0?utm_source=feedburner&amp;utm_medium=feed&amp;utm_campaign=Feed%3A+nature%2Frss%2Fcurrent+%28Nature+-+Issue%29#ref-CR8\">8<\/a><\/sup><sup>,<\/sup><sup><a href=\"https:\/\/www.nature.com\/articles\/d41586-019-01406-0?utm_source=feedburner&amp;utm_medium=feed&amp;utm_campaign=Feed%3A+nature%2Frss%2Fcurrent+%28Nature+-+Issue%29#ref-CR13\">13<\/a><\/sup>\u00a0are encouraging for the future of the field. The development of readily available broadband on-chip gain would help considerably, as would techniques that can support independent and arbitrary operations on each piece of optically encoded data, without requiring vast areas of the photonic chip. Should scalable photonic neural accelerators offering high energy efficiencies eventually emerge, we might well look back on the work of Feldmann\u00a0<i>et al.<\/i>\u00a0and others in the field as important early glimpses of the technology\u2019s promise.<\/p>\n<p>&nbsp;<\/p>\n<p>&nbsp;<\/p>\n<\/div>\n<p><span class=\"emphasis\">Nature<\/span>\u00a0<strong>569<\/strong>, 199-200 (2019)<\/p>\n<p>&nbsp;<\/p>\n<div class=\"emphasis\">doi: 10.1038\/d41586-019-01406-0<\/div>\n<\/div>\n<p>&nbsp;<\/p>\n<p>&nbsp;<\/p>\n<p>(\uc6d0\ubb38: <a href=\"https:\/\/www.nature.com\/articles\/d41586-019-01406-0?utm_source=feedburner&amp;utm_medium=feed&amp;utm_campaign=Feed%3A+nature%2Frss%2Fcurrent+%28Nature+-+Issue%29\">\uc5ec\uae30<\/a>\ub97c \ud074\ub9ad\ud558\uc138\uc694~)<\/p>\n<p>&nbsp;<\/p>\n<p>&nbsp;<\/p>\n<p>&nbsp;<\/p>\n","protected":false},"excerpt":{"rendered":"<p>&nbsp; &nbsp; Experiments show how an all-optical version of an artificial neural network \u2014 a type of artificial-intelligence system \u2014 could potentially deliver better energy<a href=\"https:\/\/biochemistry.khu.ac.kr\/lab\/?p=3485\" class=\"more-link\">(more&#8230;)<\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_monsterinsights_skip_tracking":false,"_monsterinsights_sitenote_active":false,"_monsterinsights_sitenote_note":"","_monsterinsights_sitenote_category":0,"_jetpack_newsletter_access":"","_jetpack_dont_email_post_to_subs":false,"_jetpack_newsletter_tier_id":0,"_jetpack_memberships_contains_paywalled_content":false,"_jetpack_feature_clip_id":0,"_jetpack_memberships_contains_paid_content":false,"footnotes":"","jetpack_publicize_message":"","jetpack_publicize_feature_enabled":true,"jetpack_social_post_already_shared":true,"jetpack_social_options":{"image_generator_settings":{"template":"highway","default_image_id":0,"font":"","enabled":false},"version":2},"jetpack_post_was_ever_published":false},"categories":[33,35,29,30],"tags":[],"class_list":["post-3485","post","type-post","status-publish","format-standard","hentry","category-do-biology","category-lets-do-computer-science","category-lets-do-science","category-recent-science-news"],"aioseo_notices":[],"jetpack_publicize_connections":[],"jetpack_featured_media_url":"","jetpack-related-posts":[{"id":2586,"url":"https:\/\/biochemistry.khu.ac.kr\/lab\/?p=2586","url_meta":{"origin":3485,"position":0},"title":"Deep learning beats the optical diffraction limit","author":"biochemistry","date":"January 29, 2019","format":false,"excerpt":"\u00a0 \u00a0 A deep learning approach enables up to nine bits of information to be encoded per diffraction-limited area. \u00a0 \u00a0 In our digital age, we generate an ever-increasing amount of data (terabytes per day), making its storage and long-term access increasingly challenging. Hard disk drives have become very popular\u2026","rel":"","context":"In &quot;Let's Do Computer Science!&quot;","block_context":{"text":"Let's Do Computer Science!","link":"https:\/\/biochemistry.khu.ac.kr\/lab\/?cat=35"},"img":{"alt_text":"","src":"","width":0,"height":0},"classes":[]},{"id":2956,"url":"https:\/\/biochemistry.khu.ac.kr\/lab\/?p=2956","url_meta":{"origin":3485,"position":1},"title":"The next step in making arrays of single atoms","author":"biochemistry","date":"March 27, 2019","format":false,"excerpt":"\u00a0 \u00a0 Three studies have demonstrated the cooling and trapping of single strontium and ytterbium atoms in two-dimensional arrays. Such arrays could lead to advances in atomic-clock technology and in quantum simulation and computing. \u00a0 \u00a0 The world around us is made of atoms. There are enormous numbers of them,\u2026","rel":"","context":"In &quot;Let's Do Chemistry!&quot;","block_context":{"text":"Let's Do Chemistry!","link":"https:\/\/biochemistry.khu.ac.kr\/lab\/?cat=34"},"img":{"alt_text":"","src":"","width":0,"height":0},"classes":[]},{"id":4564,"url":"https:\/\/biochemistry.khu.ac.kr\/lab\/?p=4564","url_meta":{"origin":3485,"position":2},"title":"Light trapping gets a boost","author":"biochemistry","date":"October 24, 2019","format":false,"excerpt":"\u00a0 \u00a0 The ability of structures called optical resonators to trap light is often limited by scattering of light off fabrication defects. A physical mechanism that suppresses this scattering has been reported that could lead to improved optical devices. \u00a0 \u00a0 Devices called optical resonators confine light, but for only\u2026","rel":"","context":"In &quot;Let's Do Biology!&quot;","block_context":{"text":"Let's Do Biology!","link":"https:\/\/biochemistry.khu.ac.kr\/lab\/?cat=33"},"img":{"alt_text":"","src":"","width":0,"height":0},"classes":[]},{"id":3255,"url":"https:\/\/biochemistry.khu.ac.kr\/lab\/?p=3255","url_meta":{"origin":3485,"position":3},"title":"All for one and one for all (QUANTUM OPTICS)","author":"biochemistry","date":"April 8, 2019","format":false,"excerpt":"\u00a0 \u00a0 Quantum information (QI) has become a focus of research during the past two decades, with the goal of exploiting the potentialities offered by superposition and entanglement of quantum states (1). The first hardware implementations of QI relied on quantum systems hosting clean, well-isolated two-level systems such as atoms\u2026","rel":"","context":"In &quot;Essays on Science&quot;","block_context":{"text":"Essays on Science","link":"https:\/\/biochemistry.khu.ac.kr\/lab\/?cat=32"},"img":{"alt_text":"","src":"","width":0,"height":0},"classes":[]},{"id":4092,"url":"https:\/\/biochemistry.khu.ac.kr\/lab\/?p=4092","url_meta":{"origin":3485,"position":4},"title":"Making perfectly controlled arrays of molecules at rest","author":"biochemistry","date":"September 18, 2019","format":false,"excerpt":"\u00a0 \u00a0 Since their invention in the early 1970s, optical tweezers have evolved from enabling simple manipulation to applying calibrated forces on\u2014and measuring nanometer-level displacements of\u2014optically trapped objects. Optical tweezers use laser light to create a force trap that can hold nanometer- to micrometer-sized dielectric objects (1). They can noninvasively\u2026","rel":"","context":"In &quot;Let's Do Chemistry!&quot;","block_context":{"text":"Let's Do Chemistry!","link":"https:\/\/biochemistry.khu.ac.kr\/lab\/?cat=34"},"img":{"alt_text":"","src":"","width":0,"height":0},"classes":[]},{"id":428,"url":"https:\/\/biochemistry.khu.ac.kr\/lab\/?p=428","url_meta":{"origin":3485,"position":5},"title":"Robotic assembly of artificial nanomaterials","author":"biochemistry","date":"May 30, 2018","format":false,"excerpt":"\u00a0 \u00a0 (\uc6d0\ubb38) \u00a0 \u00a0 An automated robotic system is capable of assembling 2D van der Waals heterostructures of unprecedented complexity in a timely fashion. \u00a0 The emergence of robotic automation in the workplace has unleashed a hot debate among economists about the potential impact of the robot replacing the\u2026","rel":"","context":"In &quot;Let's Do Chemistry!&quot;","block_context":{"text":"Let's Do Chemistry!","link":"https:\/\/biochemistry.khu.ac.kr\/lab\/?cat=34"},"img":{"alt_text":"","src":"","width":0,"height":0},"classes":[]}],"jetpack_sharing_enabled":false,"jetpack_shortlink":"https:\/\/wp.me\/p9Xo1j-Ud","_links":{"self":[{"href":"https:\/\/biochemistry.khu.ac.kr\/lab\/index.php?rest_route=\/wp\/v2\/posts\/3485","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/biochemistry.khu.ac.kr\/lab\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/biochemistry.khu.ac.kr\/lab\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/biochemistry.khu.ac.kr\/lab\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/biochemistry.khu.ac.kr\/lab\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=3485"}],"version-history":[{"count":1,"href":"https:\/\/biochemistry.khu.ac.kr\/lab\/index.php?rest_route=\/wp\/v2\/posts\/3485\/revisions"}],"predecessor-version":[{"id":3486,"href":"https:\/\/biochemistry.khu.ac.kr\/lab\/index.php?rest_route=\/wp\/v2\/posts\/3485\/revisions\/3486"}],"wp:attachment":[{"href":"https:\/\/biochemistry.khu.ac.kr\/lab\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=3485"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/biochemistry.khu.ac.kr\/lab\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=3485"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/biochemistry.khu.ac.kr\/lab\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=3485"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}