{"id":2091246,"date":"2022-01-29T22:40:00","date_gmt":"2022-01-29T20:40:00","guid":{"rendered":"http:\/\/nhub.news\/?p=2091246"},"modified":"2022-01-30T00:04:29","modified_gmt":"2022-01-29T22:04:29","slug":"ai-chip-startup-ceremorphic-comes-out-of-stealth-mode","status":"publish","type":"post","link":"http:\/\/nhub.news\/fr\/2022\/01\/ai-chip-startup-ceremorphic-comes-out-of-stealth-mode\/","title":{"rendered":"AI chip startup Ceremorphic comes out of stealth mode"},"content":{"rendered":"<p style=\"text-align: justify;\"><b>A host of novel chip features including analog computation will deliver a more-efficient, more reliable, more secure AI processor for multiple markets, CEO Venkat Mattel claims.<\/b><br \/>\nThe battle to change the computer industry so that machines can better compute artificial intelligence tasks, especially deep learning, continues to birth new and interesting potential future stars. On Monday, Ceremorphic of San Jose, California, formally debuted chip efforts that have been kept in a stealth mode for two years, discussing a chip the company claims will revolutionize the efficiency of A.I. computing in terms of power consumption. \u00ab\u00a0It&rsquo;s counterintuitive today, but higher performance is lower power, said Venkat Mattela, founder and CEO of the company, in an interview with ZDNet via Zoom. Mattela believes that numerous patents on low-power operation will enable his company&rsquo;s chip to produce the same accuracy on signature tasks of machine learning with much less computing effort. \u00ab\u00a0What I&rsquo;m trying to do is not just building a semiconductor chip but also the math and the algorithms to reduce the workload,\u00a0\u00bb he said. \u00ab\u00a0If a workload takes a hundred operations, I want to bring it down to fifty operations, and if fifty operations cost less energy than a hundred, I want to say mine is a higher-performance system.\u00a0\u00bb Mattela is wading into a heavily contested market, one where startups such as Cerebras Systems, Graphcore, and SambaNova have received vast sums of money, and where, for all their achievements, they still struggle to topple the industry heavyweight, Nvidia. Mattela is inclined to take the long view. His last startup, Redpine Signals, was built over a period of fourteen years, starting in 2006. That company was sold to chip maker Silicon Labs in March of 2020 for $314 million for its low-power Bluetooth and WiFi chip technology. (The chip is now being used in the recently introduced Garmin Fenix 7 smartwatch.) Also: Meta says it will soon have the world&rsquo;s fastest AI supercomputer The lesson of that seventeen-year effort at Redpine and now at Ceremorphic is twofold: \u00ab\u00a0I have a lot of patience,\u00a0\u00bb he observed of himself with a chuckle. And, \u00ab\u00a0I don&rsquo;t do incremental things.\u00a0\u00bb Mattela contends that when he takes on a problem in an area of chip design, it is in such a way as to get meaningfully ahead of the state of the art. The Redpine wireless chip technology Silicon Labs bought, he said, went up against the offerings of giant companies, Qualcomm and Broadcom in Bluetooth and WiFI. \u00ab\u00a0I took a big challenge, I went against them, but only with one metric, ultra-low-energy wireless, twenty-six times less energy than the best in the industry,\u00a0\u00bb said Mattela. Now, Mattela believes he has a similarly winning focus on power, along with three other qualities he deems both unique in the AI chip market and essential to the discipline: reliability, quantum-safe security, and an ability to function in multiple markets. To make all that possible, Mattela held onto the microprocessor assets that had been developed at Redpine, to form the foundation of Ceremorphic, and retained eighteen employees from that effort, whom he has complemented by hiring another 131 people. The company has offices in both San Jose, the official HQ, and a gleaming new office building in Hyderabad, India. Also: Cerebras continues &lsquo;absolute domination&rsquo; of high-end compute, it says, with world&rsquo;s hugest chip two-dot-oh Mattela has an intriguing list of 26 U.S. patents with his name on them, and an equally intriguing list of 14 U.S. patent applications from the last few years. What Mattela dubs a \u00ab\u00a0Hierarchical Learning Processor,\u00a0\u00bb or HLP, consists of a computing element for machine learning running at 2-gigahertz; a custom floating-point unit at the same clock frequency; a custom-designed multi-threading workload scheduling approach; and specially-designed 16-lane PCIe gen-6 circuitry to connect the processor to a system&rsquo;s host processor such as an x86 chip. The last of these, the PCIe part, could almost be its own company, claims Mattela. \u00ab\u00a0Right now, what is in production is PCIe-4, the dominant one, and PCIe-5 just started last year,\u00a0\u00bb explained MattelA. \u00ab\u00a0And with us, PCIe-6 will be in production in 2024 \u2014 I own that technology.\u00a0\u00bb \u00ab\u00a0That&rsquo;s $12 million if you had to license that,\u00a0\u00bb he said of PCIe-6. \u00ab\u00a0That alone is a significant thing to design.\u00a0\u00bb The PCIe link will allow Mattela to further refine the energy consumption of a total system, he said. At the heart of the chip&rsquo;s advantage are analog circuits resting underneath digital. Some companies have used analog circuits extensively for AI processing, the most well-known being startup Mythic, which in 2020 revealed a chip that can multiply vectors and matrices \u2014 the heart of machine learning \u2014 not as digital multiplications but as combinations of continuous energy wave forms in accordance with Ohm&rsquo;s Law, what the company calls analog computing. The Ceremorphic HLP chip will use analog computing more selectively than Mythic, Mattela told ZDNet. \u00ab\u00a0At the lowest level of the hierarchy\u00a0\u00bb of chip functionality, \u00ab\u00a0I do analog computation,\u00a0\u00bb explained Mattela. \u00ab\u00a0But higher level, I don&rsquo;t do analog because I want to make the programming model easy.\u00a0\u00bb That means \u00ab\u00a0twenty-three patterns\u00a0\u00bb for multiply-accumulate in analog via the HLP&rsquo;s micro-architecture. The analog multiplications will be a more-efficient use of voltage than digital, he argued. \u00ab\u00a0At a higher level, it looks like a vector processing, and data-path processing combination.\u00a0\u00bb The various chip features will contribute to making possible the four qualities Mattela promotes. Also: &lsquo;We are the best-funded AI startup,&rsquo; says SambaNova co-founder Olukotun following SoftBank, Intel infusion In addition to power-efficient operation, there is reliability. AI silicon has a reliability problem today, claimed Mattela. Machine learning chips have gotten vastly larger than conventional microprocessors. Nvidia&rsquo;s \u00ab\u00a0A100\u00a0\u00bb GPU is already a fairly hefty, by classic standards,826 square millimeters. But novel chips from startups can be much larger, such as Cerebras&rsquo;s WSE-2 chip, measuring 45,225 square millimeters, almost the entire surface of an eight-inch silicon wafer. \u00ab\u00a0When you have more silicon there is a greater possibility of failure, because there are alpha particles, neutron bombardment,\u00a0\u00bb observed Mattela. \u00ab\u00a0In the last two years, people are already saying, my systems are failing in the data center.\u00a0\u00bb Mattela claims a unique hardware-software combination will enable his chip to \u00ab\u00a0predict faults and correct them.\u00a0\u00bb \u00ab\u00a0Reliable performance computing engineering is our key contribution,\u00a0\u00bb he said. The third quality Mattela is emphasizing is security, including protection against future quantum systems that could conceivably break conventional data security. Also: Graphcore brings new competition to Nvidia in latest MLPerf AI benchmarks \u00ab\u00a0So far, security systems have been designed to counter hacking by humans,\u00a0\u00bb explains Mattela. \u00ab\u00a0But going forward, you can&rsquo;t assume computing power will be limited, and that it will take two days to break it [a system], you had better assume perhaps two minutes!\u00a0\u00bb The Ceremorphic chip has \u00ab\u00a0quantum-resistant random-number generation,\u00a0\u00bb said Mattela, which \u00ab\u00a0cannot be broken by a very high-performance computer.\u00a0\u00bb In practical terms, said Mattela, that means such a system would take perhaps a month to break, affording a customer time to change the security key to foil the attack. The fourth property is what Mattela refers to as scaling. What Mattela means by that is addressing multiple markets with one chip. The chip will be able to function in deep learning, but also automotive applications, in robotics, in life sciences, and in some sort of future Metaverse application. The same HLP will serve to do both training and inference, the two aspects of machine learning. Scaling to multiple markets, claimed Mattela, will make his chip more relevant than those of competitors. Startups such as Cerebras are impressive but not ultimately as relevant, he argues. \u00ab\u00a0It&rsquo;s very fine engineering, yes, and you can always do something that nobody else can do, but your purpose is not to do something which nobody does,\u00a0\u00bb said Mattela. \u00ab\u00a0Your purpose is to create an outcome that everybody makes money, and it has impact, it has some value to the market.\u00a0\u00bb Of course, Cerebras and the others are shipping product while Mattela hasn&rsquo;t even produced samples yet. To make Ceremorphic&rsquo;s design a winner, Mattela has what would appear to be an ace in his pocket: Taiwan Semi&rsquo;s 5-nanometer chip process, which is one of the manufacturing giant&rsquo;s \u00ab\u00a0advanced technologies,\u00a0\u00bb a chip process to which not every customer is given access. \u00ab\u00a0When I say to people that I am doing 5-nanometer, they say, How could you get 5-nanometer,\u00a0\u00bb said Mattela, with evident delight. \u00ab\u00a0Some of these companies with hundreds of millions in funding are not in 5-nm, they are in 7-nm.\u00a0\u00bb One reason is a close relationship Mattela cultivated with TSM years ago when he was at analog chip giant Analog Devices. More important, his sale of Redpine to Silicon Labs boosted his credibility. TSM, he says, had to believe that he can see his product through to fruition, for only then does TSM get paid in full. \u00ab\u00a0It takes many, many years,\u00a0\u00bb said Mattela, referring to the production process. \u00ab\u00a0I have to spend $200 million, I have to produce a chip, the chip has to work, and then if it works, they [TSM] will get paid at that time.\u00a0\u00bb Mattela&rsquo;s team designed the chip as a prototype initially in what&rsquo;s called a shuttle run, a small batch of chips, in the slightly less-sophisticated 7-nanometer process. The company will this year expand its shuttle run batches to 5-nanometer. While Ceremorphic expects to provide first customer samples of its chip next year, full scale production in 5-nanometer will probably not occur until 2024, he says. \u00ab\u00a0These are very aggressive dates\u00a0\u00bb for design in a leading-edge process such as 5-nano, Mattela observed of the timeline. A gating factor is cost. To move from shuttle run to full use of a wafer \u2014 what&rsquo;s known as full mask \u2014 is the difference between $2 million and over $10 million, he points out. For that, Ceremorphic will need to raise further capital. So far, Ceremorphic is self-funded, with Mattela and friends and family putting together $50 million in a Series A round, an extremely small amount of funding relative to AI chip startups such as SambaNova that have received billions in venture capital. Mattela&rsquo;s preferred route to future funding, he said, is through partnerships, though a formal Series B investment round is also a possibility in 2023. The first instantiation of the HLP will be as a PCI card that bundles together what the chip needs to function in a computer system. \u00ab\u00a0Typical system OEMs are the target,\u00a0\u00bb he said. That route to market, he believes, will make his machine more broadly available. \u00ab\u00a0Every company needs a training supercomputer,\u00a0\u00bb he said. \u00ab\u00a0I want to provide the training supercomputer that can be affordable to every enterprise.\u00a0\u00bb In contrast to something mammoth such as Facebook&rsquo;s Research SuperComputer for AI \u2014 6,080 GPUs and 175 petabytes of flash storage \u2014 the Ceremorphic PCI blade would be intended to make the technology more accessible. \u00ab\u00a0If I can provide one-tenth of a building size computer in a box, that&rsquo;s the sweet spot.\u00a0\u00bb While his part is not yet shipping, Mattela is already predicting a rapid shakeout in the AI chip startup market. The challengers such as Graphcore have raised a lot of money and made very little revenue, he speculates, just a fraction of what Nvidia makes in a quarter off of AI. \u00ab\u00a0There are four to five companies today, they have close to five billion today [in capital raised], that&rsquo;s a lot of money,\u00a0\u00bb said Mattela. But, \u00ab\u00a0the number one company, every quarter makes two billion dollars,\u00a0\u00bb referring to Nvidia&rsquo;s data center revenue. \u00ab\u00a0If you don&rsquo;t even make one percent of the number one company, and you&rsquo;re losing money, that&rsquo;s not a business,\u00a0\u00bb said Mattela. The shake-out will come sooner rather than later, Mattela prophesies, because of profligacy. \u00ab\u00a0In today&rsquo;s hot market, they went and got money, good for them, but whatever money they got, I don&rsquo;t think they really figured out how to spend the money because the amount being spent is just abnormal,\u00a0\u00bb he said. Junior engineers are being paid vast sums at the AI startups, he maintains. \u00ab\u00a0If a fresher is getting $200K [in annual salary], that&rsquo;s not sustainable,\u00a0\u00bb he said, using the tech industry jargon for the most junior position, \u00ab\u00a0because the guy will be productive after two years, but, by then, the money is already gone.\u00a0\u00bb<\/p>\n","protected":false},"excerpt":{"rendered":"<p>A host of novel chip features including analog computation will deliver a more-efficient, more reliable, more secure AI processor for multiple markets, CEO Venkat Mattel claims. The battle to change the computer industry so that machines can better compute artificial intelligence tasks, especially deep learning, continues to birth new and interesting potential future stars. On [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":2091245,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":[],"categories":[93],"tags":[],"_links":{"self":[{"href":"http:\/\/nhub.news\/fr\/wp-json\/wp\/v2\/posts\/2091246"}],"collection":[{"href":"http:\/\/nhub.news\/fr\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/nhub.news\/fr\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/nhub.news\/fr\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/nhub.news\/fr\/wp-json\/wp\/v2\/comments?post=2091246"}],"version-history":[{"count":0,"href":"http:\/\/nhub.news\/fr\/wp-json\/wp\/v2\/posts\/2091246\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"http:\/\/nhub.news\/fr\/wp-json\/wp\/v2\/media\/2091245"}],"wp:attachment":[{"href":"http:\/\/nhub.news\/fr\/wp-json\/wp\/v2\/media?parent=2091246"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/nhub.news\/fr\/wp-json\/wp\/v2\/categories?post=2091246"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/nhub.news\/fr\/wp-json\/wp\/v2\/tags?post=2091246"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}