Always Be Curious #220: Inside the Apple M4 chip, TSMC's packaging, and AI hallucinations
This week in ABC: TechInsights takes a peek under the hood of Apple M4, TSMC is doing some pioneering work in packaging, and AI can be a bit cooky.
📅 I’m unplugging from intarwebz for a bit, so the next ABC publishes Sun 7 July
This week, advanced chip packaging was once again in the spotlight with rumors about TSMC’s research into rectangular panel-like substrates. As far as rumors go, it’s not far-fetched: both Intel and Samsung have confirmed to be researching new ways of packaging chips, including panel-level packaging, and when asked for comment, TSMC said it "closely monitors progress and development in advanced packaging, including panel-level packaging." 👀
🔎 Not too long ago, chip packaging was viewed as a relatively low-tech field in the chip manufacturing ecosystem. But over the last few years, the field has boomed because of its importance to unlocking the full computational and memory power of the world’s most advanced chips used in AI applications. For example, for NVIDIA’s latest AI chip beasts like the Blackwell B200, TSMC pioneered “chip-on-wafer-on-substrate” (CoWoS) technology that makes it possible to combine two such chips and connect them with eight high bandwidth memories (HBMs). This enables fast data throughput and accelerated computing performance. But chips are packing ever more punch: they keep adding more transistors and integrating more memory. That means chips are getting bigger, so the chip package needs to grow as well. The use of rectangular (instead of round) substrates for packaging would allowing more sets of chips to be placed on each wafer. It’s not going to happen overnight though: the shift would require significant changes in facilities, robotics and automated material handling systems to be able to handle different substrate shapes and sizes. The timeline for development covers the coming decade.
☝️ If we take a chip insight here, let it be this one: when the economics work, it will happen. ✨ It doesn’t matter how irrelevant, crazy or out of reach the technology was in the past. Once technological vision and cost factor click, really, anything is possible. This is exactly what makes AI such a powerful lever, catalyst or gamechanger for many different facets of science, technology, industry, the economy. When steered with a strong ethical grounding and a fierce focus on reducing power consumption, it can unlock the next level. It is worth it. So it will happen.
Have a good week, stay safe and sound,
👨💻The round-up in sci-tech💡
🔮 Why does AI hallucinate? (MIT Technology Review)
The tendency to make things up is holding chatbots back. But that’s just what they do.
🌍 Ray Kurzweil on how AI will transform the physical world (The Economist)🔐
The changes will be particularly profound in energy, manufacturing and medicine, says the futurist.
😯 NVIDIA’s Jensen Huang is on top of the world. So why is he worried? (The Information)🔒
Huang has been around long enough to have witnessed the spectacular downfalls of hardware giants.
🛰️ Is an old NASA probe about to redraw the frontier of the solar system? (New Scientist)
The New Horizons mission to Pluto, now zooming out of the Kuiper belt, has made a discovery that could upend what we know about where the solar system ends.
🔌 AI is wreaking havoc on global power systems (Bloomberg)🔐
New artificial intelligence data centers are coming online so fast that the electricity demand is straining global power grids and threatening clean energy goals.
🪐 The dominant model of the universe is creaking (The Economist)🔐
Dark energy could break it apart.
🔮 How physics can improve image-generating AI (The Economist)🔐
The laws governing electromagnetism and even the weak nuclear force could be worth mimicking.
🥶 The enduring mystery of how water freezes (Quanta Magazine)
Making ice requires more than subzero temperatures. The unpredictable process takes microscopic scaffolding, random jiggling and often a little bit of bacteria.
☀️ Mars got cooked by a recent solar storm (The New York Times)🎁
Days after light shows filled Earth’s skies with wonder, the red planet was hit by another powerful outburst of the sun.
🤓This week in chips⚠
🔥TSMC explores radical new chip packaging approach to feed AI boom (Nikkei)
Rectangular substrates to unlock more power are also being tested by Intel and Samsung.
"The trend is certain. The size of the package will only grow bigger as chipmakers squeeze more computing power out of chips used for AI data center computing," one chip executive told Nikkei Asia. "But this is still at an early stage. For example, the coating of photoresists in cutting-edge chip packaging on a new shape of substrate is one of the bottlenecks. It takes the deep pockets of chipmakers like TSMC to push equipment makers to change equipment designs."
🍎 Introducing TSMC N3E: the power behind Apple's M4 chip (TechInsights)🔒
In a recent teardown of the Apple iPad Pro 11-inch, TechInsights revealed details of Apple's latest silicon: the Apple M4 SoC, codenamed TMRV93, built on TSMC's advanced N3E process. This surprise release demonstrates Apple's agility in adopting cutting-edge semiconductor technologies ahead of schedule.
🔥 Monolithic CFET devices with stacked bottom and top contacts (imec)
At the 2024 IEEE Symposium on VLSI Technology & Circuits, imec presented electrically functional CMOS CFET devices with stacked bottom and top source/drain contacts. Read my backgrounder for context around the tiny transistors that will power the next decade.
🇺🇸 Intel delivers leading-edge foundry node with Intel 3 technology; on path back to process leadership (Intel)
Intel shared details of the Intel 3 process node, which is their ultimate FinFET-based node, at the annual VLSI Symposium.
📈 Training compute of frontier AI models grows by 4-5x per year (EpochAI)
Epoch’s expanded AI model database shows that the compute used to train recent models grew 4-5x yearly from 2010 to May 2024. They find similar growth in frontier models, recent large language models, and models from leading companies.
🔥 Cerebras, maker of wafer-scale AI chips, prepares for IPO (Techzine)
Cerebras is promising that its huge AI chips are continuing Moore's Law. Is it a viable NVIDIA alternative?
📈By the numbers📉
🤪💰 NVIDIA rips past entire stock-market values in Germany, France (MarketWatch)
Skyrocketing shares of NVIDIA have lifted the U.S. company to a massive market value exceeding entire country stock markets in Europe, according to Deutsche Bank.
❤️For the love of tech❤️
I just wanted to take a moment to mark the birthday of graphic artist M.C. Escher in appreciation of his genius and incredible legacy. ✍️🎈 Born on June 17, 1898, Escher's mind-bending creations continue to fascinate people around the world. His intricate designs and impossible architectures challenge our perceptions and invite us into a world where the boundaries of reality and illusion blur. Go see his work at the permanent exhibition in The Hague!
🎈Permanent exhibition | Museum Escher in The Palace
The exhibition has over 120 prints.
Always Be Curious is the personal newsletter of Sander Hofman, Senior Creative Content Strategist at ASML. Opinions expressed in this curated newsletter are my own and do not necessarily reflect those of my employer.
Enjoy your time off, Sander. Another great edition as always.
Here's an awesome documentary on Escher that I absolutely love, check it out :)
https://youtu.be/bNjUR1Nn710?si=6EkHMwcqX5f1aN5I