Always Be Curious #265: Wafer-scale chips, HBM8, and Nintendo’s success
This week in ABC: Wafer-scale accelerators have a use-case, the HBM roadmap goes all the way to HBM8, and the Switch 2 is selling like hot cakes
📅 Due to an upcoming holiday break, the next ABC will publish Sun 6 July
Sup Curious Clan! ✌️ This week, it’s clear that the Nintendo Switch 2 is selling like hot cakes and I'm pretty fascinated by the silicon choices for this console. 🔥 This thing basically crams desktop-class gaming into a mobile form factor through interesting and clever chip design.
To pull it off, Nintendo looked to NVIDIA. The GPU giant became an essential part of the Switch 2 story. Because they didn't just hand Nintendo a best-in-class-but-off-the-shelf mobile chip. Instead, they put in 1,000 engineering-years to build fully customized silicon specifically for Nintendo's requirements: desktop-class graphics performance within handheld thermal and power constraints. NVIDIA essentially redesigned their RTX architecture for this exact use case, delivering a 10x increase in graphical performance over the original Switch. 📈
The onboard GPU is built on the Ampere architecture and has specialized processing units called RT Cores. These beasts handle ray tracing math - essentially shooting millions of virtual light rays through a 3D scene to calculate realistic lighting, reflections, and shadows. Each core can process ray-triangle intersections at massive parallel scale. This used to require desktop GPUs pulling 200+ watts. Now it's running on battery power in your hands.
Then there are the Tensor Cores, which handle the matrix math for deep learning inference. The Switch 2 is running neural networks live to upscale 1080p gameplay to 4K when docked. The chip predicts what higher-resolution pixels should look like based on lower-res input. It's kinda like computational photography, but for gaming.
Alright, enough gawking. Are they back in stock yet? 🤪
Have a good week, stay safe and sound,
👨💻The round-up in sci-tech💡
🔭 How astronomers will deal with 60 million billion bytes of imagery (The New York Times)🎁
The Vera C. Rubin Observatory will make the study of stars and galaxies more like the big data-sorting exercises of contemporary genetics and particle physics.
🧑🔬The radical idea that space-time remembers could upend cosmology (New Scientist)
There are new hints that the fabric of space-time may be made of "memory cells" that record the whole history of the universe. If true, it could explain the nature of dark matter and much more
🔮 Can AI quicken the pace of math discovery? (The New York Times)🎁
Breakthroughs in pure mathematics can take decades. A new Defense Department initiative aims to speed things up using artificial intelligence.
📷 Why JPEG became the Web's favorite image format (IEEE Spectrum)
What makes JPEG files so special? Discover the technical magic that keeps them at the forefront of digital photography.
📖 How a 30-year-old techno-thriller predicted our digital isolation (MIT Technology Review)
In Irwin Winkler’s proto–cyber thriller, a woman’s most reliable companion is the screen in front of her.
🎭 Real risk to youth mental health is ‘addictive use,’ not screen time alone, study finds (The New York Times)🎁
Researchers found children with highly addictive use of phones, video games or social media were two to three times as likely to have thoughts of suicide or to harm themselves.
Seriously, what is ‘superintelligence’? (WIRED)
In this episode of “Uncanny Valley,” we talk about Meta’s recent investment in Scale AI and its move to build a superintelligence AI research lab. So we ask: What is superintelligence anyway?
🤓This week in chips⚠
🔥 Wafer-scale accelerators could redefine AI (UC Riverside)
The promise of a new type of computer chip that could reshape the future of artificial intelligence and be more environmentally friendly is explored in a technology review paper published by UC Riverside engineers in the journal Device.
🏗️ Next-gen HBM architecture detailed including HBM4, HBM5, HBM6, HBM7 & HBM8: up to 64 TB/s bandwidth, 240 GB capacity per 24-Hi Stack, embedded cooling (wccftech)
Next-generation HBM standards, including HBM4, HBM5, HBM6, HBM7 & HBM8, have been detailed, offering big advances in each generation.
📊 Tales from NVIDIA's backyard (Claus Aasholm)
How deep are the roots of the AI revolution? The CapEx increases, but Nvidia's stock price is flatlining. How strong is the business and its prospects.
🍿 Moore’s Law scaling roadmaps to 2039 (TechTechPotato)
“Explore IMEC's latest semiconductor roadmap, revealing the future of process node technology through 2039, including advancements in transistor design like FinFETs, nanosheets, and the revolutionary complimentary FETs (CFETs). Discover the cutting-edge lithography machines, from current EUV to future High-NA EUV and even "hyper NA EUV," essential for enabling these smaller, more powerful chips. We'll also delve into the complex challenges of power delivery and the integration of exotic 2D materials, offering a rare glimpse into the monumental engineering feats required for tomorrow's silicon.”
🇹🇼Taiwan imposes technology export controls on Huawei, SMIC (Bloomberg)🔐
Taiwan has blacklisted Huawei Technologies Co. and Semiconductor Manufacturing International Corp., dealing another major blow to the two companies spearheading China’s efforts to develop cutting-edge AI chip technologies.
💰 Texas Instruments injects $60B into US chip manufacturing (TechTarget)
Texas Instruments unveils a $60B plan to expand chip manufacturing across Texas and Utah, creating 60,000 jobs through partnerships with tech giants.
📉 Wolfspeed in rough waters, European rivals stay the course (Bits&Chips)
While their American competitor threatens to collapse, Infineon and STMicroelectronics stay the course with a diversified product portfolio and their China-for-China strategies.
🇮🇪 Ireland's “Silicon Island” strategy could boost Europe’s analog ecosystem (eenews Europe)
Ireland's “Silicon Island” strategy could boost Europe’s analog ecosystem. The strategy is aimed at advancing analog, digital, and packaging capabilities.
🔥 Photonic processor could streamline 6G wireless signal processing (MIT)🔐
MIT researchers developed a photonic AI hardware accelerator designed specifically to handle wireless signal processing, reducing latency. Their architecture encodes and processes data using light to dramatically accelerate deep learning computations on an edge device.
📈By the numbers📉
📈 Billion dollar carbon nanotube market continues to grow (IDTechEx)
IDTechEx Research Article: Following decades of hype and false dawns, nanocarbons are now beginning to see substantial market adoption and growth. This article delves into the continued growth of the carbon nanotube market, which is set to exceed $1.25 billion by 20235. IDTechEx assesses the key applications driving the growth, and the leading players poised to take a significant market share.
❤️For the love of tech❤️
💉 Nanoneedle patch offers painless alternative to traditional cancer biopsies (Phys.org)
A patch containing tens of millions of microscopic nanoneedles could soon replace traditional biopsies, scientists have found. The patch offers a painless and less invasive alternative for millions of patients worldwide who undergo biopsies each year to detect and monitor diseases like cancer and Alzheimer's. The research is published in Nature Nanotechnology.
🕹️ Nintendo Switch 2 review: bigger and better, for a higher price (The New York Times)🎁
The hardware improvements in the new $450 Nintendo will make this an automatic upgrade for fans of the two-in-one console.
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.