<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Tapa Ghosh on Medium]]></title>
        <description><![CDATA[Stories by Tapa Ghosh on Medium]]></description>
        <link>https://medium.com/@sixsamuraisoldier?source=rss-4e3a17edde2f------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*jWGp6Je68e_XwlNwgQBrXg.jpeg</url>
            <title>Stories by Tapa Ghosh on Medium</title>
            <link>https://medium.com/@sixsamuraisoldier?source=rss-4e3a17edde2f------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 12:26:45 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@sixsamuraisoldier/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Real2Sim]]></title>
            <link>https://medium.com/@sixsamuraisoldier/real2sim-4dd98fd85edf?source=rss-4e3a17edde2f------2</link>
            <guid isPermaLink="false">https://medium.com/p/4dd98fd85edf</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[robotics]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[self-driving-cars]]></category>
            <category><![CDATA[robots]]></category>
            <dc:creator><![CDATA[Tapa Ghosh]]></dc:creator>
            <pubDate>Wed, 21 Oct 2020 02:58:24 GMT</pubDate>
            <atom:updated>2020-10-21T03:00:38.226Z</atom:updated>
            <content:encoded><![CDATA[<p>Autonomous robots have long been a mainstay of science fiction, but the problem has proven to be remarkably hostile to AI systems.</p><p>A big part of the problem is the ambiguity and variation in perceiving the real world for manipulation tasks. To that end, AI systems can perform much better in simulated environments, and a lot of work has gone into transferring promising results from simulations to real world environments, so-called “Sim2Real”.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/850/0*oGbDy17mNI_gkmih.png" /><figcaption>From Sim2Real</figcaption></figure><p>However, this line of research has been challenging, see for example, the comic below:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*op71FGSNhkU3p7Tq.jpg" /><figcaption>from Dileep George’s AGIcomics</figcaption></figure><p>The joke here is that it’s easier to try and make reality look like simulations rather than it is to try and get better at transferring simulations to reality.</p><p>However, I will argue that recent advancements in LIDAR technology will enable the joke of “Real2Sim” to become real and integral to robotics progress.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*yded3kdcFeIxxD1_.jpg" /><figcaption>Velodyne HDL-64 from Mapix Technologies</figcaption></figure><p>Historically, LIDAR has been used for robotics applications such as autonomous vehicles and general robotic manipulation.</p><p>For example:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/0*5psRulRsBlCkgpo3.png" /><figcaption>From the Robot Report</figcaption></figure><p>It’s pretty obvious that the clump of dots that are read out by today’s LIDAR systems are nowhere near what is necessary to be useful for solving the perception stack fully, and is nowhere near “Real2Sim”.</p><p>However, advancements in LIDAR systems, driven by perceived demand from automotive has resulted in short-range LIDAR that is qualitatively different from the LIDAR of old.</p><p>Short range LIDAR has improved tremendously, leveraging advances in flash LIDAR technology to achieve much high resolutions than previously possible.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ofzp-mNS-cEXp3Q5.jpg" /><figcaption>From Sense Photonics</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/801/0*5gurii48zBj4dQva" /><figcaption>Apple’s VCSEL+SPAD based LIDAR, From Patently Apple</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/850/0*qAVwnTuxKZDjswBD.png" /><figcaption>From Leafcutter LIDAR Project</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/705/0*Zf5pbwUl35olB_zl.jpg" /><figcaption>Ouster Multi Point Flash LIDAR</figcaption></figure><p>Such high resolution LIDAR systems will result in robots being able to perceive the world in a new way, more akin to viewing a simulation than reality, hence “Real2Sim”. Just as MEMS sensors were the “peace dividends of the smartphone wars” that enabled quadcopters, so too will LIDAR enable many robotics applications to become a reality.</p><p>Finally, looking more towards the future, newer LIDAR systems are smaller than ever, and despite the packaging expensive of silicon photonics, integrated LIDAR systems will be cheap and <strong><em>small enough to embed everywhere, even on the “fingertips” and other surfaces of robotic manipulators, enabling something even more powerful than a sense of touch.</em></strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*SZYhIn42WS3Irznc.jpeg" /><figcaption>DARPA “LIDAR on a Chip” Project</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/730/0*Nvaw9m_Iw2GnbpZm" /><figcaption>Integrated LIDAR “on a chip”, from Voyant Photonics</figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4dd98fd85edf" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[pCells: Recent Support from Supersoscillation Researh]]></title>
            <link>https://medium.com/@sixsamuraisoldier/pcells-recent-support-from-supersoscillation-researh-e0f18a7b352d?source=rss-4e3a17edde2f------2</link>
            <guid isPermaLink="false">https://medium.com/p/e0f18a7b352d</guid>
            <category><![CDATA[cell-phones]]></category>
            <category><![CDATA[5g]]></category>
            <category><![CDATA[electromagnetic]]></category>
            <category><![CDATA[networking]]></category>
            <category><![CDATA[telecommunication]]></category>
            <dc:creator><![CDATA[Tapa Ghosh]]></dc:creator>
            <pubDate>Wed, 31 Jul 2019 17:46:22 GMT</pubDate>
            <atom:updated>2019-07-31T17:46:22.945Z</atom:updated>
            <content:encoded><![CDATA[<p>Some of you may have heard of pCells from Artemis Networks, a technology proposed by Artemis Networks that claims to nearly arbitrarily increase wireless channel capacity given a fixedamount of spectrum.</p><p>Initially, the concept was met with deep skepticism.</p><p>This may be a good case of where experts can be wrong, even about deeply technical subjects.</p><p>Just a couple of examples:</p><p><a href="https://www.youtube.com/watch?v=5bO0tjAdOIw">https://www.youtube.com/watch?v=5bO0tjAdOIw</a> (there is a loud naysayer in the audience)</p><p><a href="https://www.theregister.co.uk/2011/08/01/dido_snake_oil_or_saviour/">https://www.theregister.co.uk/2011/08/01/dido_snake_oil_or_saviour/</a></p><p><a href="https://news.ycombinator.com/item?id=7732821">https://news.ycombinator.com/item?id=7732821</a></p><p><a href="https://www.quora.com/Is-pCell-from-Artemis-really-the-Holy-Grail-of-wireless-networking">https://www.quora.com/Is-pCell-from-Artemis-really-the-Holy-Grail-of-wireless-networking</a></p><p>At first glance it would appear that this would violate the Shannon limit, but, rules are after all, meant to be broken. Or in this case, followed exactly but just sidestepped for our specific application ;)</p><p>Here, I show that recent results in research in the area of superoscilations support the generation of sub-diffraction limit balls of RF radiation.</p><p>The first paper shows that sub-diffraction limit three dimensional balls of RF radiation can be created, which is similar to the Artemis claim.</p><p><a href="https://journals.aps.org/prb/abstract/10.1103/PhysRevB.95.075148">Broadband superoscillation brings a wave into perfect three-dimensional focus</a></p><p>However, superoscillations must, in order to avoid violating physical laws, come at the price of energy dumped into side-lobes, which are of much higher power than the actual desired focusing region. Even if these side-lobes were dumped away from the transmitter, due to the polynomial increase in side-band power and exponential increase in variability, this would result in the tremendous RF noise pollution.</p><p>However, it turns out that this is only true in free space, when we consider superoscillations only in a fixed region, it turns out that sub-diffraction superoscillation focusing points without side lobes are a valid solution to Maxwell’s equations in an enclosed region.</p><p><a href="https://www.nature.com/articles/srep08449">Superoscillations without Sidebands: Power-Efficient Sub-Diffraction Imaging with Propagating Waves</a></p><p>This has recently been experimentally demonstrated in a so-called “Huygens Box”</p><p><a href="https://arxiv.org/abs/1810.05998">Active Huygens&#39; Box: Metasurface-Enabled Arbitrary Electromagnetic Wave Generation Inside a Cavity</a></p><p>Outside of the research context of superoscillations, this technique has been known as “distributed beamforming” and has been investigated for a wide array of applications, from wireless energy transfer to interception-free communication:</p><p><a href="https://dl.acm.org/citation.cfm?id=3214268">Energy-Ball</a></p><iframe src="https://drive.google.com/viewerng/viewer?url=https%3A//pdfs.semanticscholar.org/ba20/ee9057cb97bcb4db80153edbc4484b1bd758.pdf&amp;embedded=true" width="600" height="780" frameborder="0" scrolling="no"><a href="https://medium.com/media/e261a34513aa8402c05c08148b7b6fd6/href">https://medium.com/media/e261a34513aa8402c05c08148b7b6fd6/href</a></iframe><p><a href="https://www.hindawi.com/journals/mpe/2016/7426429/">A Novel Three-Dimensional Beamforming Antenna Array for Wireless Power Focusing</a></p><p>Now, the concept is not without limitations, as it is still ultimately limited by the number of transmitting antennae. In addition, the concept appears infeasible for wireless power transfer given the extremely low total energy to projected energy ratios of the order of ~100–1000X. However, considering the enormous complexity and cost of 5G and mmWave 5G systems, this may be a possible way forward for telecommunications, especially given the backwards compatibility with 4G phones. At the very least, it validates Artemis’s claims and negates the skepticism of “industry experts”. Furthermore, possibilities exist to dramatically extend the focusing capability, at least in the regime of information and imaging, which I plan to publish in the future.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e0f18a7b352d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why GaAs and III-Vs Probably Won’t Replace Silicon]]></title>
            <link>https://medium.com/@sixsamuraisoldier/why-gaas-and-iii-vs-probably-wont-replace-silicon-568b4701901d?source=rss-4e3a17edde2f------2</link>
            <guid isPermaLink="false">https://medium.com/p/568b4701901d</guid>
            <category><![CDATA[moores-law]]></category>
            <category><![CDATA[semiconductors]]></category>
            <category><![CDATA[materials]]></category>
            <category><![CDATA[cpu]]></category>
            <category><![CDATA[processors]]></category>
            <dc:creator><![CDATA[Tapa Ghosh]]></dc:creator>
            <pubDate>Fri, 28 Dec 2018 15:01:01 GMT</pubDate>
            <atom:updated>2018-12-28T15:01:01.341Z</atom:updated>
            <content:encoded><![CDATA[<p>How many times have you heard that in the future we’ll ditch silicon and transition to Gallium Arsenide* and our chips will be super fast?</p><p>*or Indium Gallium Arsenide (InGaAs), Germanium or some other high carrier mobility material with a lower bandgap than silicon.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KaHMh592VQiAfigIMhR6tw.png" /></figure><p>As the “End of Moore’s Law” is nigh, this has become a somewhat hot topic. Although I mentioned, GaAs and III-Vs, I’ll talk about compound semiconductors and germanium in general.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FWazlTFoNjiksM5G8tXUbQ.png" /><figcaption>Source: IEEE Spectrum</figcaption></figure><p>So, naturally, I thought I’d write my first blog post with a rather pessimistic view.</p><p>In my opinion, there are five major problems with the use of III-Vs</p><ol><li><strong>Cost</strong></li><li><strong>Native Oxides (or lack thereof)</strong></li><li><strong>pMOS/nMOS tradeoff</strong></li><li><strong>Materials challenges (“a thousand papercuts”)</strong></li><li><strong>Performance (!) &lt; — A new one</strong></li></ol><h4>Cost</h4><p>Moore’s Law, and the semiconductor world in general, is as much about economics as it is about technology, and the two are very intertwined as the cost of technology limits the access people have to it.</p><p>Historically, cost has been one of the major barriers against the use of III-Vs.</p><p>Unlike silicon wafers, which are produced in vast volumes with extraordinary purity and at large sizes (300mm), III-V crystals are produced in much smaller volumes and are thus far more expensive, 200mm III-V wafers can cost <a href="https://news.stanford.edu/pr/2015/pr-chips-laser-lift-032415.html">as much as $5,000</a> while <strong>300</strong> mm silicon wafers cost less than $100.</p><p>And unfortunately, III-V wafers tend to use the Bridgman-Stockbarger (or Bridgman method for short) growth process, which doesn’t allow us to easily reuse the Czochralski process silicon wafer factories either.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*MwWpdMENKskGQOJr.png" /><figcaption>Czochralski Process- Picture Courtesy Wikipedia</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/400/1*itsBqaCpYQ7YSp0RkOHyaw.png" /><figcaption>Bridgman Technique- Picture Courtesy Wikipedia</figcaption></figure><p>Making larger wafers like 300mm/12-inch wafers with high yield is also not a trivial process for III-Vs.</p><p>However, in recent years the ante has shifted to re-using silicon wafers for the bulk, and either bonding a thin slice of a III-V wafer on top of a mostly silicon wafer or by growing a III-V layer on the silicon. This works especially well in “XOI” type configurations: X on insulator (usually an oxide) where X is the semiconductor of choice, say a III-V.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tDTuP--BzO9E-EWcm_eyqA.png" /><figcaption>Example Epi growth of III-Vs on silicon (<a href="http://www.ee.ust.hk/~ptc/Papers/2017/qli_Review2017.pdf">source</a>)</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/610/0*PCPfBxQnq45haWPt.jpg" /><figcaption>Ion Cut SOI Wafer Manufacturing Process (<a href="https://www.researchgate.net/figure/Ion-Cut-smart-cut-process-stacking-single-crystal-silicon-Soitec_fig1_301099496">source</a>)</figcaption></figure><p>So, in my opinion, the cost issue is on its way to being resolved. Challenges remain, as they always do, but IMHO, they are not the deal breaker.</p><h4>Native Oxides</h4><p>Silicon also has one major advantage, a stable and easy to form native oxide (silicon dioxide).</p><p>Most III-Vs either lack a native oxide or the native oxide (e.g. GeOx) is unstable and hard to grow or deposit. Furthermore, it often causes damage to the material, e.g. Fermi level pinning at the interface (also was a challenge in the transition to high-k dielectrics in silicon, which was partially mitigated by transition layers and metal gates).</p><p>Historically, this was a major argument used against the use of III-Vs, however, as we have advanced along the transistor development curve, we have moved away from standard SiO2 to high-k dielectrics in silicon MOSFETs as well to reduce gate leakage. As a result, the native oxide argument is less convincing now.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/550/0*_SmkS-jlU4f3qrLL.jpg" /><figcaption>Source: AnandTech</figcaption></figure><p>Furthermore, oxide growth techniques on III-Vs have significantly improved, and the use of quantum well FETs (QW-FETs) has been proposed, so IMHO the native oxide problem is quite solvable as well.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/478/0*wTVR2iebK5nH9x0k" /><figcaption>QWFETs (<a href="https://www.realworldtech.com/intel-10nm-qwfet/">Source</a>)</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/594/0*pQ0-z9NyIim5cZAy" /><figcaption>QW-FinFET (Intel, IEDM 2011)</figcaption></figure><h4>pMOS/nMOS Tradeoff</h4><p>III-Vs make excellent nMOS transistors due to their very high electron mobility, but most III-Vs actually possess <em>worse </em>hole mobility than silicon.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/380/0*zHCVpj2qihDe9IG1.png" /><figcaption>Electrons vs Holes (<a href="https://www.allaboutcircuits.com/textbook/semiconductors/chpt-2/electrons-and-holes/">source</a>)</figcaption></figure><p>For CMOS, you need both, nMOS (electrons) and pMOS (holes). Historically, this has been a major challenge.</p><p>Fortunately however, there is a way around this: use Germanium for pMOS transistors (extremely high hole mobility, higher than silicon’s electron mobility) and use III-Vs for nMOS transistors.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_Mu8ST2Gfvf6iSj1eENh0Q.png" /><figcaption>Germanium and III-V pMOS/nMOS (<a href="https://www.semanticscholar.org/paper/Ge%2FIII-V-MOS-device-technologies-for-low-power-Takagi-Takenaka/a8250d5cf25711a0d370a679d866688533ac9c58">source</a>)</figcaption></figure><p>The cost of this of course, is massively increased complexity, but it can likely be done, especially if someone like TSMC puts their considerable resources behind it.</p><h4>Materials Challenges</h4><p>One other thing to realize is that silicon is one of the best known materials to science, second only to perhaps steel.</p><p>There are many “unknown unknowns” here, such as say, thermal conductivity. Since we’re using a silicon bulk for the wafer, chances are that won’t really matter much, but might it in odd cases?</p><p>One more thing to worry about could be the impurities in III-Vs, which tend to be high compared to silicon. That will get better over time, but will it be easy? Probably not.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/400/0*wIYrkLrsJUJyttT7.gif" /><figcaption>Thermal Conductivity of Si vs Ge (<a href="http://www.iue.tuwien.ac.at/phd/palankovski/node34.html">source</a>)</figcaption></figure><p>Another potential source of danger is the mechanical properties of the material. Although this seems irrelevant for electronics, there are many areas where it’s relevant. For example, advanced packaging like 3D stacking is used more and more where the mechanicals do in fact matter.</p><p>Now, at this point I’m sure you’re thinking that these are all surmountable problems and daddy wants his 12 GHz processor, so hurry up and please move to GaAs.</p><p>However, it turns out that soon silicon may *outperform III-Vs*.</p><h4>Performance</h4><p>This is probably the most important section.</p><p>Basically, at some point, say 2nm Lg (gate length), there is a phenomenon known as Direct Source to Drain Tunneling. Source-drain tunneling results in severely degraded transistor performance and is arguably a fundamental limitation to transistor scaling. Of course, this has been known to some extent for some time, see below:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_xXwpvwWckaN7EyBThQ1nA.png" /><figcaption><a href="http://www.monolithic3d.com/blog/is-there-a-fundamental-limit-to-miniaturizing-cmos-transistors1">Source</a></figcaption></figure><p>To understand how this applies to III-Vs, it’s useful to look at the equation for the S-D tunneling probability. Notably, the probability is roughly exponentially dependent upon the the effective carrier mass. So, as we reduce the effective carrier mass by using a high mobility material, such as a III-V, the probability of S-D tunneling shoots up exponentially.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/419/0*XnYPj8jCU55pNYsl" /><figcaption>Equation for S-D Tunneling (<a href="http://www.monolithic3d.com/blog/is-there-a-fundamental-limit-to-miniaturizing-cmos-transistors1">source</a>)</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6M7rBuo1FpEdi9O7qTgolQ.png" /><figcaption>(<a href="http://www.ispd.cc/slides/2016/3_1.pdf">source</a>)</figcaption></figure><p>So, basically, what we end up with is that at lower geometry nodes we actually want a <strong><em>high</em></strong> effective carrier mass.</p><p>This phenomenon has been explored in a couple of papers, such as <a href="https://ieeexplore.ieee.org/document/7479214">Transistor design for 5nm and beyond: Slowing down electrons to speed up transistors </a>and <a href="https://ieeexplore.ieee.org/document/6523122/authors#authors">Engineering Nanowire n-MOSFETs at Lg &lt; 8nm</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3KwGQ9Ez-6JEBnPeFYatrQ.png" /><figcaption><a href="http://www.ispd.cc/slides/2016/3_1.pdf">Source</a></figcaption></figure><p><strong>So unfortunately, it looks like although most of the major historical problems with III-Vs have been solved or have become solvable, now we have reached a point in which III-Vs are no longer advantageous.</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=568b4701901d" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>