@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20240408

Episode Date: April 8, 2024

- TSMC Arizona Back on Schedule - Dutch Government Support for ASML - Intel Foundry profitability - TSMC expands CoWoS capacity - SK hynix $3.9B for Indiana HBM Fab and R&D with Purdue - Quantinuum and Microsoft show major improvement for qubit fidelity [audio mp3="https://orionx.net/wp-content/uploads/2024/04/HPCNB_20240408.mp3"][/audio] The post HPC News Bytes – 20240408 appeared first on OrionX.net.

Welcome to HPC News Bites, a weekly show about important news in the world of supercomputing, AI, and other advanced technologies. Hi, everyone. Welcome to HPC News Bites. I'm Doug Black. Hi, Shaheen. Chips are in the news, as they are nearly every week. There's a positive report from a Taiwanese publication called Money.UDN that

TSMC's Arizona advanced chip plant is, quote, sprinting towards trial production this month and aims to go into full production by year's end or the first half of 2025. Also, the Dutch government has sweetened the pot for chip manufacturing equipment maker ASML, which had threatened to explore a move to other countries. But now the Netherlands government will spend nearly $3 billion to strengthen the business climate in the region where ASML is located. And finally, Intel released financial numbers for its foundry business, the expansion of which CEO Pat Gelsinger says will be a big part of their strategy moving forward. But the numbers as of now are not stellar, a loss of nearly $7 billion last year,

and Gelsinger says it won't break even for another three years. ASML, of course, is the only company in the world that makes EUV and high-NA EUV machines that are needed for high-end chips from TSMC, Samsung, and Intel. There are more producers for the older technology, DUV, standing for deep ultraviolet. ASML made extreme ultraviolet, EUV, work and then upped the ante with high numerical aperture, INA EUV, and it has become Europe's most valuable technology company as a result. It makes sense for the Dutch and EU governments to make sure they maintain the lead. When you see the global competitive environment with governments pouring hundreds of billions of dollars to boost their

home team, and the importance of chips for national security, and the fact that the home team in the US is playing catch-up, then you have renewed appreciation for the context. The message is you cannot entrust a must-have, like national security, to markets even when they are operating efficiently, let alone when your competition is investing heavily. Intel is pushing hard and meeting milestones and needs all our support with a realistic, holistic, and long-term view. Staying with chips, we mentioned the new word COWOS, C-O-W-O-S, some time ago and expected to hear it more often. Well, we have. TSMC has increased its investment and ordered additional equipment as it builds more COWOS capability, which signals growing demand and importance of packaging technology. CoBOS, by the way, stands for chip on wafer on substrate.

It is a so-called 2.5D packaging technology that puts multiple chiplets next to each other, each bonding via micro bumps with a large interconnect layer below. That's called a silicon interposer. That's the chip-on-wafer part. The WASP part is thinning the interposer layer to expose connections to the package substrate layer below and further treat it until it is ready to bond. That forms the chips that we see with a ball grid array underneath ready to connect to the circuit board. Other packaging technologies you might have heard are InfoPOP, standing for integrated fan out package on package, is a wafer level system integration

technology that Apple uses in most of its products, and SOIC, which stands for small outline integrated circuit, which is an industry standard for low cost solutions and many applications. Part of this is TSMC's plan to develop what they call their Advanced Backend FAB 7, the sixth such plan to go online in August 2025 in southern Taiwan. This will increase the production for GPUs to 32,000 wafers per month by the end of 2024 and 44,000 a month by the end of 2025. And by the way, this is not to the exclusion of the other packaging technologies that also will be growing to accommodate different market segments. Shaheen, it seems the rate of change, innovation, and resulting chip performance

keeps going through the roof. If we accept as a sort of Moore's law corollary that in the end we're talking about the doubling of system performance every 18 months, then the contributions of GPUs and chiplets and co-was is driving performance improvement at really an incredible rate. More on chips. Memory chip company SK Hynix announced it is investing nearly $4 billion in a high bandwidth memory fab in Indiana and an R&D center with Purdue University. The project will be an advanced semiconductor production line that will produce next generation HBM used in GPUs for AI. So SK Hynix is a major South Korean company and it plans to begin mass production in the second

half of 2028, while the new facility will also develop future generation of chips and house an advanced packaging R&D line. Finally, there's Quantum News that's creating quite a stir. A Continuum-Microsoft joint effort has achieved an advance in fault-tolerant quantum by, quote, demonstrating the most reliable logical qubits, an achievement previously believed to be years from realization, unquote, demonstrating the most reliable logical qubits, an achievement previously believed to be years from realization, unquote, the companies said. Yeah, fidelity and therefore quantum error correction is the most important need for

qubits and the specific operations that they can perform. Late last year, there was a major advance by QERA and Harvard, and this new result ups the ante. They've demonstrated error rates that are 800 times lower than corresponding physical error rates and created enough redundant qubits to form four logical qubits, which lets them run 14,000 error-free instances of a quantum circuit. So on the one hand, this is very big progress, but on the other, you're still talking very few qubits. But it certainly accelerates the path towards a quantum computer that can deflect errors

and keep running what the industry calls a fault-tolerant universal quantum computer. All right, that's it for this episode. Thank you all for being with us. HPC Newsbytes is a production of OrionX in association with InsideHPC. Shaheen Khan and Doug Black host the show. Every episode is featured on InsideHPC.com and posted on OrionX.net. Thank you for listening.