An AI friendly fabįrom the get-go, Rapidus is focusing on three main markets: serving those building traditional CPUs and GPUs those designing edge compute and thirdly, and most importantly, AI chip startups that don't yet have the volume to make sense for TSMC. Regardless of whether these technologies prove a competitive advantage, he says, capacity constraints alone should be more than enough to guarantee Rapidus' success, he opines. Richard was hesitant on sharing details on the specific technologies Rapidus is currently exploring beyond its work on process tech with IBM, but says more details would be coming later this year.

Emerging and maturing technologies like backside power delivery, advanced packaging, and next-gen substrates are expected to play a bigger role in driving performance and efficiency than transistor density in the future. It's no secret that performance or efficiency gains from process shrinks, something that we used to take for granted, have become far less impressive as Moore's Law has slowed. 'We might come to an inflection point where, if you just judge a fab, based on when a node is available for mass production, it's a very partial view of the competitiveness of the offering.' However, process tech isn't the only consideration these days, he argues. This puts Rapidus at least a year behind its competition, a fact Richard acknowledges. In fact, Intel expects to ship internal chips built using its 20A node - its 2nm equivalent - later this year, followed by its mass market 18A node next year. TSMC, Samsung, and Intel are slated to being shipping their 2nm process tech in 2025 with volume production expected for 2026. One of the company's biggest challenges relates to timing. However, achieving this goal may not be so easy as it faces stiff competition from the likes of Samsung, TSMC, and Intel. Rapidus aims to establish itself as an alternative source of advanced semiconductor manufacturing. Early contract wins included Canadian RISC-V chip dev Tenstorrent, which plans to produce its AI accelerator in Rapidus' fabs. Most recently, Rapidus opened an office in Silicon Valley which is focused on sales, business development, marketing, and tech support for potential clients.

'We are on our way to have close to 200 engineers in Albany that are working on the technology and preparing the transfer,' Richard says, adding that pilot production at the Hokkaido plant is set to begin in 2026 followed by volume production in 2027. Following installation, Rapidus will begin the technology transfer of IBMs process tech, being developed in parallel at its research facility in Albany, New York. It began construction of its IIM-1 plant in Hokkaido in September, which will produce wafers based on an advanced 2nm process node developed in collaboration with IBM.Īrmed with $6.1 billion in funding from the Japanese government, Richard says the foundry biz is on track to complete construction later this year and begin installing equipment early next. If you're not familiar with Rapidus, the Japanese foundry upstart is a relatively new player having gotten its start in 2022. 'If you believe that the market is going to move from the previous forecasts of maybe $80 billion dollars at the advanced node, to $150 billion, it doesn't take a large market share to make Rapidus a great success,' Richard adds.

Chained echoes