{"id":28236,"date":"2019-08-15T17:11:42","date_gmt":"2019-08-15T17:11:42","guid":{"rendered":"https:\/\/silvaco.com\/%eb%b6%84%eb%a5%98%eb%90%98%ec%a7%80-%ec%95%8a%ec%9d%8c\/advanced-materials-and-new-architectures-for-ai-applications\/"},"modified":"2019-08-15T17:11:42","modified_gmt":"2019-08-15T17:11:42","slug":"advanced-materials-and-new-architectures-for-ai-applications","status":"publish","type":"post","link":"https:\/\/silvaco.com\/ko\/tcad-ko\/tcad-blogs-ko\/advanced-materials-and-new-architectures-for-ai-applications\/","title":{"rendered":"Advanced Materials and New Architectures for AI Applications"},"content":{"rendered":"<div id='template_overview'  class='avia-section main_color avia-section-small avia-no-border-styling  avia-bg-style-scroll  avia-builder-el-0  avia-builder-el-no-sibling   container_wrap fullsize' style='background-color: #ffffff;  margin-top:0px; margin-bottom:0px; '  ><div class='container' ><main  role=\"main\" itemprop=\"mainContentOfPage\"  class='template-page content  av-content-full alpha units'><div class='post-entry post-entry-type-page post-entry-28236'><div class='entry-content-wrapper clearfix'>\n<div class='flex_column_table av-equal-height-column-flextable -flextable' style='margin-top:20px; margin-bottom:0px; '><div class=\"flex_column av_three_fourth  flex_column_table_cell av-equal-height-column av-align-top first  avia-builder-el-1  el_before_av_one_fourth  avia-builder-el-first  \" style='padding:0px 0px 0px 0px ; border-radius:0px; '><section class=\"av_textblock_section \"  itemscope=\"itemscope\" itemtype=\"https:\/\/schema.org\/BlogPosting\" itemprop=\"blogPost\" ><div class='avia_textblock  '   itemprop=\"text\" ><h1>Advanced Materials and New Architectures for AI Applications<\/h1>\n<p style=\"box-sizing: border-box; margin: 0px 0px 20px; line-height: 24px; letter-spacing: normal; color: #0a0a0a; font-family: 'Open Sans', sans-serif; font-size: 14px; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: 400; orphans: 2; text-align: left; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; background-color: #ffffff; text-decoration-style: initial; text-decoration-color: initial;\">*** This article by Tom Dillinger was first published on\u00a0<a href=\"https:\/\/semiwiki.com\/eda\/silvaco\/7770-advanced-materials-and-new-architectures-for-ai-applications\/\" target=\"_blank\" rel=\"noreferrer noopener\">SemiWiki.com<\/a>\u00a0***<\/p>\n<p style=\"box-sizing: border-box; margin: 0px 0px 20px; line-height: 24px; letter-spacing: normal; color: #0a0a0a; font-family: 'Open Sans', sans-serif; font-size: 14px; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: 400; orphans: 2; text-align: left; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; background-color: #ffffff; text-decoration-style: initial; text-decoration-color: initial;\">Over the past 50 years in our industry, there have been three invariant principles:<\/p>\n<ul style=\"box-sizing: border-box; margin-top: 0px; margin-bottom: 1rem; color: #0a0a0a; font-family: 'Open Sans', sans-serif; font-size: 14px; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; orphans: 2; text-align: left; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; background-color: #ffffff; text-decoration-style: initial; text-decoration-color: initial;\">\n<li>Moore\u2019s Law drives the pace of Si technology scaling<\/li>\n<li>system memory utilizes MOS devices (for SRAM and DRAM)<\/li>\n<li>computation relies upon the \u201cvon Neumann\u201d architecture<\/li>\n<\/ul>\n<div class=\"wp-block-image\" style=\"box-sizing: border-box; margin-bottom: 1em; color: #0a0a0a; font-family: 'Open Sans', sans-serif; font-size: 14px; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; orphans: 2; text-align: left; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; background-color: #ffffff; text-decoration-style: initial; text-decoration-color: initial;\">\n<figure class=\"alignleft\">\n<div id=\"attachment_8566\" style=\"width: 406px\" class=\"wp-caption aligncenter\"><img loading=\"lazy\" decoding=\"async\" aria-describedby=\"caption-attachment-8566\" class=\"wp-image-8566 size-full\" src=\"\/wp-content\/uploads\/2020\/04\/Blog35.jpg\" alt=\"\" width=\"396\" height=\"280\" srcset=\"https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog35.jpg 396w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog35-300x212.jpg 300w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog35-260x185.jpg 260w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog35-43x30.jpg 43w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog35-63x45.jpg 63w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog35-48x34.jpg 48w\" sizes=\"(max-width: 396px) 100vw, 396px\" \/><p id=\"caption-attachment-8566\" class=\"wp-caption-text\">Block diagram of the von Neumann architecture<\/p><\/div><\/p>\n<p style=\"box-sizing: border-box; margin: 0px 0px 20px; line-height: 24px; letter-spacing: normal; color: #0a0a0a; font-family: 'Open Sans', sans-serif; font-size: 14px; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: 400; orphans: 2; text-align: left; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; background-color: #ffffff; text-decoration-style: initial; text-decoration-color: initial;\">Today, all three tenets are being challenged \u2013 our industry is facing disruptive changes across the board.<\/p>\n<p style=\"box-sizing: border-box; margin: 0px 0px 20px; line-height: 24px; letter-spacing: normal; color: #0a0a0a; font-family: 'Open Sans', sans-serif; font-size: 14px; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: 400; orphans: 2; text-align: left; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; background-color: #ffffff; text-decoration-style: initial; text-decoration-color: initial;\">Although Moore\u2019s Law clearly still has a sustaining technology roadmap for another process node or two, the related costs are a driving force behind novel packaging technologies providing (heterogeneous) multi-die integration \u2013 some refer to this direction as \u201cmore than Moore\u201d.<\/p>\n<p style=\"box-sizing: border-box; margin: 0px 0px 20px; line-height: 24px; letter-spacing: normal; color: #0a0a0a; font-family: 'Open Sans', sans-serif; font-size: 14px; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: 400; orphans: 2; text-align: left; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; background-color: #ffffff; text-decoration-style: initial; text-decoration-color: initial;\">Traditional MOS system memory processes are being challenged by new approaches. Resistive RAM (ReRAM) and magneto-resistive RAM (MRAM) technologies utilize unique materials to alter the electrical (or magnetic) properties of a layer between two electrodes during a set\/reset memory write cycle. The subsequent memory read cycle senses the electrical resistance through the layer to determine the bitcell value.<\/p>\n<p style=\"box-sizing: border-box; margin: 0px 0px 20px; line-height: 24px; letter-spacing: normal; color: #0a0a0a; font-family: 'Open Sans', sans-serif; font-size: 14px; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: 400; orphans: 2; text-align: left; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; background-color: #ffffff; text-decoration-style: initial; text-decoration-color: initial;\">And, lastly, the von Neumann architecture for computation is being called into question, as to whether an alternative method could provide improved power\/performance measures for certain applications \u2013 especially, AI machine learning tasks. The latency and power dissipation associated with frequent data\/instruction memory access is an issue, especially for (massively) parallel and pipelined algorithms. An alternative architecture is needed for this class of computation to optimize power efficiency.<\/p>\n<p style=\"box-sizing: border-box; margin: 0px 0px 20px; line-height: 24px; letter-spacing: normal; color: #0a0a0a; font-family: 'Open Sans', sans-serif; font-size: 14px; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: 400; orphans: 2; text-align: left; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; background-color: #ffffff; text-decoration-style: initial; text-decoration-color: initial;\">At the recent Silvaco Users Global event (SURGE), Blessy Alexander, Director of Design Technology at Applied Materials gave a compelling keynote presentation,\u201cRole of Connective Materials to Systems for AI\u201d. Her talk provided insights into how memory technology and computation architecture changes are needed to address the rapidly growing applications for neural network training and inference.<\/p>\n<div id=\"attachment_8567\" style=\"width: 495px\" class=\"wp-caption aligncenter\"><img loading=\"lazy\" decoding=\"async\" aria-describedby=\"caption-attachment-8567\" class=\"wp-image-8567 size-full\" src=\"\/wp-content\/uploads\/2020\/04\/Blog36.jpg\" alt=\"\" width=\"485\" height=\"270\" srcset=\"https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog36.jpg 485w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog36-300x167.jpg 300w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog36-43x24.jpg 43w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog36-63x35.jpg 63w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog36-48x27.jpg 48w\" sizes=\"(max-width: 485px) 100vw, 485px\" \/><p id=\"caption-attachment-8567\" class=\"wp-caption-text\">Neural network architecture<\/p><\/div>\n<p style=\"box-sizing: border-box; margin: 0px 0px 20px; line-height: 24px; letter-spacing: normal; color: #0a0a0a; font-family: 'Open Sans', sans-serif; font-size: 14px; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: 400; orphans: 2; text-align: left; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; background-color: #ffffff; text-decoration-style: initial; text-decoration-color: initial;\">Recall that neural networks are comprised of multiple \u201clayers\u201d of nodes, where the computation at each node is commonly a weighted multiply-accumulate among the inputs to the node. An AI application would partition a data frame into neural network input values, and attempt to classify the data based upon the final results of the neural network computation. Each node in a layer consists of weighted multiply-add computations, whose total is refined by an activation function to provide the node output value. An initial training sequence is used to establish the weighting factors throughout the neural network layers, to achieve high classification accuracy on the reference dataset. In inference mode, the network is presented with runtime data to perform the classification evaluation.<\/p>\n<p style=\"box-sizing: border-box; margin: 0px 0px 20px; line-height: 24px; letter-spacing: normal; color: #0a0a0a; font-family: 'Open Sans', sans-serif; font-size: 14px; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: 400; orphans: 2; text-align: left; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; background-color: #ffffff; text-decoration-style: initial; text-decoration-color: initial;\">The implementation of the NN using a von Neumann architecture requires accessing data and weight values from (traditional) memory, performing the computation in a (traditional) CPU, and storing the node results back in memory \u2013 an inefficient and highly dissipative approach.<\/p>\n<p style=\"box-sizing: border-box; margin: 0px 0px 20px; line-height: 24px; letter-spacing: normal; color: #0a0a0a; font-family: 'Open Sans', sans-serif; font-size: 14px; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: 400; orphans: 2; text-align: left; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; background-color: #ffffff; text-decoration-style: initial; text-decoration-color: initial;\">Blessy highlighted the activity currently underway to realize neural networks using GPU hardware and\/or custom chip designs \u2013 optimizations distinct from a von Neumann implementation. Then, she challenged our traditional thinking further, suggesting that the ReRAM technologies offer a\u00a0<strong>very<\/strong>\u00a0unique opportunity to represent the weight values, and thus, impact the entire NN computation.<\/p>\n<div id=\"attachment_8568\" style=\"width: 630px\" class=\"wp-caption aligncenter\"><img loading=\"lazy\" decoding=\"async\" aria-describedby=\"caption-attachment-8568\" class=\"wp-image-8568 size-full\" src=\"\/wp-content\/uploads\/2020\/04\/Blog37.jpg\" alt=\"\" width=\"620\" height=\"141\" srcset=\"https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog37.jpg 620w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog37-300x68.jpg 300w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog37-43x10.jpg 43w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog37-63x14.jpg 63w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog37-48x11.jpg 48w\" sizes=\"(max-width: 620px) 100vw, 620px\" \/><p id=\"caption-attachment-8568\" class=\"wp-caption-text\">Oxide-based Resistive RAM (OxRAM) \u2014 a cross section of the dielectric layer, illustrating ion\/vacancy motion (From Perniola, \u201cOXRAM Memories: A Disruptive Technology for Disruptive Designs\u201d)<\/p><\/div>\n<p style=\"box-sizing: border-box; margin: 0px 0px 20px; line-height: 24px; letter-spacing: normal; color: #0a0a0a; font-family: 'Open Sans', sans-serif; font-size: 14px; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: 400; orphans: 2; text-align: left; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; background-color: #ffffff; text-decoration-style: initial; text-decoration-color: initial;\">For example, consider the novel ReRAM technology known as Oxide-based resistive RAM, or OxRAM for short. For each bit cell, a unique dielectric material resides between two terminals. During a write cycle, the density of oxide vacancies is altered (i.e., dielectric locations where an oxygen atom has been removed), corresponding to a \u201cset\u201d or \u201creset\u201d value. A read cycle measures the effective resistance of the cell, due to the motion of oxygen ions and vacancies (analogous to the motion of electrons and holes in a semiconductor).<\/p>\n<p style=\"box-sizing: border-box; margin: 0px 0px 20px; line-height: 24px; letter-spacing: normal; color: #0a0a0a; font-family: 'Open Sans', sans-serif; font-size: 14px; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: 400; orphans: 2; text-align: left; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; background-color: #ffffff; text-decoration-style: initial; text-decoration-color: initial;\">The figures below illustrate the resistive characteristics of the OxRAM technology, with a depiction of a \u201c1T-1R\u201d bit cell in a traditional memory array configuration, with sense amplifiers used to detect each specific cell\u2019s electrical properties during a read access.<\/p>\n<p><div id=\"attachment_8569\" style=\"width: 386px\" class=\"wp-caption aligncenter\"><img loading=\"lazy\" decoding=\"async\" aria-describedby=\"caption-attachment-8569\" class=\"wp-image-8569 size-full\" src=\"\/wp-content\/uploads\/2020\/04\/Blog38.jpg\" alt=\"\" width=\"376\" height=\"280\" srcset=\"https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog38.jpg 376w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog38-300x223.jpg 300w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog38-43x32.jpg 43w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog38-63x47.jpg 63w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog38-48x36.jpg 48w\" sizes=\"(max-width: 376px) 100vw, 376px\" \/><p id=\"caption-attachment-8569\" class=\"wp-caption-text\">Figure 4. OxRAM cell resistance \u2014 Monte Carlo process simulation results (From Aziza, et al, \u201cOxide based resistive RAM: ON\/OFF resistance analysis versus circuit variability\u201d, IEEE 2014 International Symposium on Defect and Fault Tolerance in VLSI and Nanotechnology Systems.)<\/p><\/div><br \/>\n<div id=\"attachment_8563\" style=\"width: 397px\" class=\"wp-caption aligncenter\"><img loading=\"lazy\" decoding=\"async\" aria-describedby=\"caption-attachment-8563\" class=\"wp-image-8563 size-full\" src=\"\/wp-content\/uploads\/2020\/04\/Blog39.jpg\" alt=\"\" width=\"387\" height=\"280\" srcset=\"https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog39.jpg 387w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog39-300x217.jpg 300w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog39-43x31.jpg 43w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog39-63x46.jpg 63w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog39-48x35.jpg 48w\" sizes=\"(max-width: 387px) 100vw, 387px\" \/><p id=\"caption-attachment-8563\" class=\"wp-caption-text\">OxRAM based \u201c1T 1R\u201d memory configuration (From Aziza, et al.)<\/p><\/div><\/p>\n<p><em style=\"box-sizing: border-box; color: #0a0a0a; font-family: 'Open Sans', sans-serif; font-size: 14px; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; orphans: 2; text-align: left; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; background-color: #ffffff; text-decoration-style: initial; text-decoration-color: initial;\">\u201cBut, what if the resistive characteristics of the ReRAM cell technology could be used to directly represent the weights of a neural network node?\u201d<\/em><span style=\"color: #0a0a0a; font-family: 'Open Sans', sans-serif; font-size: 14px; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; orphans: 2; text-align: left; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; background-color: #ffffff; text-decoration-style: initial; text-decoration-color: initial; display: inline !important; float: none;\">, Blessy asked.<\/span><\/p>\n<div id=\"attachment_8564\" style=\"width: 570px\" class=\"wp-caption aligncenter\"><img loading=\"lazy\" decoding=\"async\" aria-describedby=\"caption-attachment-8564\" class=\"wp-image-8564 size-full\" src=\"\/wp-content\/uploads\/2020\/04\/Blog40.jpg\" alt=\"\" width=\"560\" height=\"244\" srcset=\"https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog40.jpg 560w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog40-300x131.jpg 300w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog40-43x19.jpg 43w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog40-63x27.jpg 63w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog40-48x21.jpg 48w\" sizes=\"(max-width: 560px) 100vw, 560px\" \/><p id=\"caption-attachment-8564\" class=\"wp-caption-text\">ReRAM cell resistance used as a neural network weight factor<\/p><\/div>\n<p style=\"box-sizing: border-box; margin: 0px 0px 20px; line-height: 24px; letter-spacing: normal; color: #0a0a0a; font-family: 'Open Sans', sans-serif; font-size: 14px; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: 400; orphans: 2; text-align: left; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; background-color: #ffffff; text-decoration-style: initial; text-decoration-color: initial;\">The sense amplifiers would be replaced by an accumulation + activation function. As a result, the capacity and bandwidth constraints of the von Neumann approach would be greatly alleviated. A very interesting idea.<\/p>\n<p style=\"box-sizing: border-box; margin: 0px 0px 20px; line-height: 24px; letter-spacing: normal; color: #0a0a0a; font-family: 'Open Sans', sans-serif; font-size: 14px; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: 400; orphans: 2; text-align: left; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; background-color: #ffffff; text-decoration-style: initial; text-decoration-color: initial;\">(In some regards, this would be similar to a binary-weighted summing digital-to-analog converter, I guess. Multiple ReRAM bit cells in parallel could be written to provide any of a large set of weights.)<\/p>\n<p style=\"box-sizing: border-box; margin: 0px 0px 20px; line-height: 24px; letter-spacing: normal; color: #0a0a0a; font-family: 'Open Sans', sans-serif; font-size: 14px; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: 400; orphans: 2; text-align: left; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; background-color: #ffffff; text-decoration-style: initial; text-decoration-color: initial;\">The link between AMAT and Silvaco TCAD tools was the final part of Blessy\u2019s presentation. New materials used in process technologies such as ReRAM require a close interaction among process development, physical characterization, model generation, and circuit simulation \u2014 part of the \u201cPhysical Proof-of-Concept\u201d (PPOC) development phase. Additionally, the unique system requirements of AI applications motivate new architectures, which ultimately drive into the technology and materials research \u2013 the \u201cSystems Proof-of-Concept\u201d (SPOC) phase. The ultimate success of a new system approach requires a collaborative methodology between these PPOC and SPOC development teams.<\/p>\n<div id=\"attachment_8565\" style=\"width: 458px\" class=\"wp-caption aligncenter\"><img loading=\"lazy\" decoding=\"async\" aria-describedby=\"caption-attachment-8565\" class=\"wp-image-8565 size-full\" src=\"\/wp-content\/uploads\/2020\/04\/Blog41.jpg\" alt=\"\" width=\"448\" height=\"248\" srcset=\"https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog41.jpg 448w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog41-300x166.jpg 300w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog41-43x24.jpg 43w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog41-63x35.jpg 63w, https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/Blog41-48x27.jpg 48w\" sizes=\"(max-width: 448px) 100vw, 448px\" \/><p id=\"caption-attachment-8565\" class=\"wp-caption-text\">Materials, TCAD, and System co-optimization, using a \u201cPhysical Proof-of-Concept\u201d and \u201cSystems Proof-of-Concept\u201d diagram<\/p><\/div>\n<p style=\"box-sizing: border-box; margin: 0px 0px 20px; line-height: 24px; letter-spacing: normal; color: #0a0a0a; font-family: 'Open Sans', sans-serif; font-size: 14px; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: 400; orphans: 2; text-align: left; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; background-color: #ffffff; text-decoration-style: initial; text-decoration-color: initial;\">The power\/performance\/cost of neural network computation will define how quickly the emerging AI applications will be adopted, in automotive, medical, industrial, aerospace, and consumer markets. It will require re-evaluation of some of the long-standing principles of our industry. The introduction of new memory technologies and the willingness to explore non-traditional architectures offers an interesting development opportunity.<\/p>\n<p style=\"box-sizing: border-box; margin: 0px 0px 20px; line-height: 24px; letter-spacing: normal; color: #0a0a0a; font-family: 'Open Sans', sans-serif; font-size: 14px; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: 400; orphans: 2; text-align: left; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; background-color: #ffffff; text-decoration-style: initial; text-decoration-color: initial;\">-chipguy<\/p>\n<p style=\"box-sizing: border-box; margin: 0px 0px 20px; line-height: 24px; letter-spacing: normal; color: #0a0a0a; font-family: 'Open Sans', sans-serif; font-size: 14px; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: 400; orphans: 2; text-align: left; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-text-stroke-width: 0px; background-color: #ffffff; text-decoration-style: initial; text-decoration-color: initial;\">To read the rest of this blog continue to\u00a0<a href=\"https:\/\/semiwiki.com\/eda\/silvaco\/7770-advanced-materials-and-new-architectures-for-ai-applications\/\" target=\"_blank\" rel=\"noreferrer noopener\">Advanced Materials and New Architectures for AI Applications on SemiWiki.com<\/a>.<\/p>\n<\/figure>\n<\/div>\n<\/div><\/section><\/div><div class='av-flex-placeholder'><\/div><div class=\"flex_column av_one_fourth  flex_column_table_cell av-equal-height-column av-align-top av-zero-column-padding   avia-builder-el-3  el_after_av_three_fourth  avia-builder-el-last  \" style='border-radius:0px; '><div  class='avia-builder-widget-area clearfix  avia-builder-el-4  avia-builder-el-no-sibling '><div id=\"nav_menu-27\" class=\"widget clearfix widget_nav_menu\"><div class=\"menu-blog-side-menu-korean-container\"><ul id=\"menu-blog-side-menu-korean\" class=\"menu\"><li id=\"menu-item-25033\" class=\"menu-item menu-item-type-post_type menu-item-object-page menu-item-has-children menu-item-25033\"><a href=\"https:\/\/silvaco.com\/ko\/?page_id=24362\">\ube14\ub85c\uadf8<\/a>\n<ul class=\"sub-menu\">\n\t<li id=\"menu-item-25034\" class=\"menu-item menu-item-type-post_type menu-item-object-page menu-item-25034\"><a href=\"https:\/\/silvaco.com\/ko\/corporate\/blogs\/analog-custom-ic-design-blogs\/\">Analog Custom IC Design Blogs<\/a><\/li>\n\t<li id=\"menu-item-25035\" class=\"menu-item menu-item-type-post_type menu-item-object-page menu-item-25035\"><a href=\"https:\/\/silvaco.com\/ko\/blogs\/design-ip\/\">Design IP Blogs<\/a><\/li>\n\t<li id=\"menu-item-25036\" class=\"menu-item menu-item-type-post_type menu-item-object-page menu-item-25036\"><a href=\"https:\/\/silvaco.com\/ko\/meet-silvaco-blogs\/\">Meet Silvaco Blogs<\/a><\/li>\n\t<li id=\"menu-item-25037\" class=\"menu-item menu-item-type-post_type menu-item-object-page menu-item-25037\"><a href=\"https:\/\/silvaco.com\/ko\/tcad-blogs-2\/\">TCAD Blogs<\/a><\/li>\n<\/ul>\n<\/li>\n<\/ul><\/div><\/div><\/div><\/div><\/div><!--close column table wrapper. Autoclose: 1 -->\n","protected":false},"excerpt":{"rendered":"<p>Over the past 50 years in our industry, there have been three invariant principles:<\/p>\n<p>Moore\u2019s Law drives the pace of Si technology scaling<br \/>\nsystem memory utilizes MOS devices (for SRAM and DRAM)<br \/>\ncomputation relies upon the \u201cvon Neumann\u201d architecture<\/p>\n","protected":false},"author":5,"featured_media":20265,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[7614],"tags":[],"yoast_head":"<!-- This site is optimized with the Yoast SEO Premium plugin v24.0 (Yoast SEO v24.0) - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>Advanced Materials and New Architectures for AI Applications - Silvaco<\/title>\n<meta name=\"description\" content=\"Advanced Materials and New Architectures for AI Applications\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/silvaco.com\/ko\/tcad-ko\/tcad-blogs-ko\/advanced-materials-and-new-architectures-for-ai-applications\/\" \/>\n<meta property=\"og:locale\" content=\"ko_KR\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Advanced Materials and New Architectures for AI Applications\" \/>\n<meta property=\"og:description\" content=\"Advanced Materials and New Architectures for AI Applications\" \/>\n<meta property=\"og:url\" content=\"https:\/\/silvaco.com\/ko\/tcad-ko\/tcad-blogs-ko\/advanced-materials-and-new-architectures-for-ai-applications\/\" \/>\n<meta property=\"og:site_name\" content=\"Silvaco\" \/>\n<meta property=\"article:publisher\" content=\"https:\/\/www.facebook.com\/SilvacoSoftware\/\" \/>\n<meta property=\"article:published_time\" content=\"2019-08-15T17:11:42+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/tcad_process_device-11.jpg\" \/>\n\t<meta property=\"og:image:width\" content=\"500\" \/>\n\t<meta property=\"og:image:height\" content=\"500\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/jpeg\" \/>\n<meta name=\"author\" content=\"Ingrid Schwarz\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:creator\" content=\"@SilvacoSoftware\" \/>\n<meta name=\"twitter:site\" content=\"@SilvacoSoftware\" \/>\n<meta name=\"twitter:label1\" content=\"\uae00\uc4f4\uc774\" \/>\n\t<meta name=\"twitter:data1\" content=\"Ingrid Schwarz\" \/>\n\t<meta name=\"twitter:label2\" content=\"\uc608\uc0c1 \ub418\ub294 \ud310\ub3c5 \uc2dc\uac04\" \/>\n\t<meta name=\"twitter:data2\" content=\"7\ubd84\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/silvaco.com\/ko\/tcad-ko\/tcad-blogs-ko\/advanced-materials-and-new-architectures-for-ai-applications\/\",\"url\":\"https:\/\/silvaco.com\/ko\/tcad-ko\/tcad-blogs-ko\/advanced-materials-and-new-architectures-for-ai-applications\/\",\"name\":\"Advanced Materials and New Architectures for AI Applications - Silvaco\",\"isPartOf\":{\"@id\":\"https:\/\/silvaco.com\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/silvaco.com\/ko\/tcad-ko\/tcad-blogs-ko\/advanced-materials-and-new-architectures-for-ai-applications\/#primaryimage\"},\"image\":{\"@id\":\"https:\/\/silvaco.com\/ko\/tcad-ko\/tcad-blogs-ko\/advanced-materials-and-new-architectures-for-ai-applications\/#primaryimage\"},\"thumbnailUrl\":\"https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/tcad_process_device-11.jpg\",\"datePublished\":\"2019-08-15T17:11:42+00:00\",\"dateModified\":\"2019-08-15T17:11:42+00:00\",\"author\":{\"@id\":\"https:\/\/silvaco.com\/#\/schema\/person\/f85f880bc8a7ce823b36a9072e1f0a29\"},\"description\":\"Advanced Materials and New Architectures for AI Applications\",\"breadcrumb\":{\"@id\":\"https:\/\/silvaco.com\/ko\/tcad-ko\/tcad-blogs-ko\/advanced-materials-and-new-architectures-for-ai-applications\/#breadcrumb\"},\"inLanguage\":\"ko-KR\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/silvaco.com\/ko\/tcad-ko\/tcad-blogs-ko\/advanced-materials-and-new-architectures-for-ai-applications\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"ko-KR\",\"@id\":\"https:\/\/silvaco.com\/ko\/tcad-ko\/tcad-blogs-ko\/advanced-materials-and-new-architectures-for-ai-applications\/#primaryimage\",\"url\":\"https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/tcad_process_device-11.jpg\",\"contentUrl\":\"https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/tcad_process_device-11.jpg\",\"width\":500,\"height\":500},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/silvaco.com\/ko\/tcad-ko\/tcad-blogs-ko\/advanced-materials-and-new-architectures-for-ai-applications\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\/\/silvaco.com\/ko\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Advanced Materials and New Architectures for AI Applications\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/silvaco.com\/#website\",\"url\":\"https:\/\/silvaco.com\/\",\"name\":\"Silvaco\",\"description\":\"\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/silvaco.com\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"ko-KR\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/silvaco.com\/#\/schema\/person\/f85f880bc8a7ce823b36a9072e1f0a29\",\"name\":\"Ingrid Schwarz\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"ko-KR\",\"@id\":\"https:\/\/silvaco.com\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/secure.gravatar.com\/avatar\/d3bb75be07e3e2b63d0be26dc8912ffe?s=96&d=blank&r=g\",\"contentUrl\":\"https:\/\/secure.gravatar.com\/avatar\/d3bb75be07e3e2b63d0be26dc8912ffe?s=96&d=blank&r=g\",\"caption\":\"Ingrid Schwarz\"},\"url\":\"https:\/\/silvaco.com\/ko\/author\/ingrid\/\"}]}<\/script>\n<!-- \/ Yoast SEO Premium plugin. -->","yoast_head_json":{"title":"Advanced Materials and New Architectures for AI Applications - Silvaco","description":"Advanced Materials and New Architectures for AI Applications","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/silvaco.com\/ko\/tcad-ko\/tcad-blogs-ko\/advanced-materials-and-new-architectures-for-ai-applications\/","og_locale":"ko_KR","og_type":"article","og_title":"Advanced Materials and New Architectures for AI Applications","og_description":"Advanced Materials and New Architectures for AI Applications","og_url":"https:\/\/silvaco.com\/ko\/tcad-ko\/tcad-blogs-ko\/advanced-materials-and-new-architectures-for-ai-applications\/","og_site_name":"Silvaco","article_publisher":"https:\/\/www.facebook.com\/SilvacoSoftware\/","article_published_time":"2019-08-15T17:11:42+00:00","og_image":[{"width":500,"height":500,"url":"https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/tcad_process_device-11.jpg","type":"image\/jpeg"}],"author":"Ingrid Schwarz","twitter_card":"summary_large_image","twitter_creator":"@SilvacoSoftware","twitter_site":"@SilvacoSoftware","twitter_misc":{"\uae00\uc4f4\uc774":"Ingrid Schwarz","\uc608\uc0c1 \ub418\ub294 \ud310\ub3c5 \uc2dc\uac04":"7\ubd84"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/silvaco.com\/ko\/tcad-ko\/tcad-blogs-ko\/advanced-materials-and-new-architectures-for-ai-applications\/","url":"https:\/\/silvaco.com\/ko\/tcad-ko\/tcad-blogs-ko\/advanced-materials-and-new-architectures-for-ai-applications\/","name":"Advanced Materials and New Architectures for AI Applications - Silvaco","isPartOf":{"@id":"https:\/\/silvaco.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/silvaco.com\/ko\/tcad-ko\/tcad-blogs-ko\/advanced-materials-and-new-architectures-for-ai-applications\/#primaryimage"},"image":{"@id":"https:\/\/silvaco.com\/ko\/tcad-ko\/tcad-blogs-ko\/advanced-materials-and-new-architectures-for-ai-applications\/#primaryimage"},"thumbnailUrl":"https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/tcad_process_device-11.jpg","datePublished":"2019-08-15T17:11:42+00:00","dateModified":"2019-08-15T17:11:42+00:00","author":{"@id":"https:\/\/silvaco.com\/#\/schema\/person\/f85f880bc8a7ce823b36a9072e1f0a29"},"description":"Advanced Materials and New Architectures for AI Applications","breadcrumb":{"@id":"https:\/\/silvaco.com\/ko\/tcad-ko\/tcad-blogs-ko\/advanced-materials-and-new-architectures-for-ai-applications\/#breadcrumb"},"inLanguage":"ko-KR","potentialAction":[{"@type":"ReadAction","target":["https:\/\/silvaco.com\/ko\/tcad-ko\/tcad-blogs-ko\/advanced-materials-and-new-architectures-for-ai-applications\/"]}]},{"@type":"ImageObject","inLanguage":"ko-KR","@id":"https:\/\/silvaco.com\/ko\/tcad-ko\/tcad-blogs-ko\/advanced-materials-and-new-architectures-for-ai-applications\/#primaryimage","url":"https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/tcad_process_device-11.jpg","contentUrl":"https:\/\/silvaco.com\/wp-content\/uploads\/2020\/04\/tcad_process_device-11.jpg","width":500,"height":500},{"@type":"BreadcrumbList","@id":"https:\/\/silvaco.com\/ko\/tcad-ko\/tcad-blogs-ko\/advanced-materials-and-new-architectures-for-ai-applications\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/silvaco.com\/ko\/"},{"@type":"ListItem","position":2,"name":"Advanced Materials and New Architectures for AI Applications"}]},{"@type":"WebSite","@id":"https:\/\/silvaco.com\/#website","url":"https:\/\/silvaco.com\/","name":"Silvaco","description":"","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/silvaco.com\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"ko-KR"},{"@type":"Person","@id":"https:\/\/silvaco.com\/#\/schema\/person\/f85f880bc8a7ce823b36a9072e1f0a29","name":"Ingrid Schwarz","image":{"@type":"ImageObject","inLanguage":"ko-KR","@id":"https:\/\/silvaco.com\/#\/schema\/person\/image\/","url":"https:\/\/secure.gravatar.com\/avatar\/d3bb75be07e3e2b63d0be26dc8912ffe?s=96&d=blank&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/d3bb75be07e3e2b63d0be26dc8912ffe?s=96&d=blank&r=g","caption":"Ingrid Schwarz"},"url":"https:\/\/silvaco.com\/ko\/author\/ingrid\/"}]}},"_links":{"self":[{"href":"https:\/\/silvaco.com\/ko\/wp-json\/wp\/v2\/posts\/28236"}],"collection":[{"href":"https:\/\/silvaco.com\/ko\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/silvaco.com\/ko\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/silvaco.com\/ko\/wp-json\/wp\/v2\/users\/5"}],"replies":[{"embeddable":true,"href":"https:\/\/silvaco.com\/ko\/wp-json\/wp\/v2\/comments?post=28236"}],"version-history":[{"count":0,"href":"https:\/\/silvaco.com\/ko\/wp-json\/wp\/v2\/posts\/28236\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/silvaco.com\/ko\/wp-json\/wp\/v2\/media\/20265"}],"wp:attachment":[{"href":"https:\/\/silvaco.com\/ko\/wp-json\/wp\/v2\/media?parent=28236"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/silvaco.com\/ko\/wp-json\/wp\/v2\/categories?post=28236"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/silvaco.com\/ko\/wp-json\/wp\/v2\/tags?post=28236"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}