Lin Xu

I've just read Aravind Ravichandran '𝗟𝗮𝘀𝘁 𝗪𝗲𝗲𝗸 𝗶𝗻 𝗘𝗮𝗿𝘁𝗵 𝗢𝗯𝘀𝗲𝗿𝘃𝗮𝘁𝗶𝗼𝗻' and I noticed Planet has announced it will be integrating an NVIDIA AI processer in one of its upcoming satellites to 'power intelligent imaging and faster insights for customers' - one to follow, notably when it comes to edge computing 👀 I also read that Canada has announced its 2030 Nature Strategy in which satellite data will play an integral part - The 2030 Nature Strategy includes 23 targets and implementation plans of which six targets include the need of satellite data 🛰️ 𝙈𝙤𝙧𝙚 𝘽𝙚𝙡𝙤𝙬: - 𝗧𝗮𝗿𝗴𝗲𝘁 𝟭: 𝗦𝗽𝗮𝘁𝗶𝗮𝗹 𝗣𝗹𝗮𝗻𝗻𝗶𝗻𝗴 𝗮𝗻𝗱 𝗘𝗳𝗳𝗲𝗰𝘁𝗶𝘃𝗲 𝗠𝗮𝗻𝗮𝗴𝗲𝗺𝗲𝗻𝘁 - 𝗧𝗮𝗿𝗴𝗲𝘁 𝟰: 𝗦𝗽𝗲𝗰𝗶𝗲𝘀 𝗥𝗲𝗰𝗼𝘃𝗲𝗿𝘆 - 𝗧𝗮𝗿𝗴𝗲𝘁 𝟳: 𝗣𝗼𝗹𝗹𝘂𝘁𝗶𝗼𝗻 𝗮𝗻𝗱 𝗕𝗶𝗼𝗱𝗶𝘃𝗲𝗿𝘀𝗶𝘁𝘆 - 𝗧𝗮𝗿𝗴𝗲𝘁 𝟴: 𝗖𝗹𝗶𝗺𝗮𝘁𝗲 𝗖𝗵𝗮𝗻𝗴𝗲 𝗮𝗻𝗱 𝗕𝗶𝗼𝗱𝗶𝘃𝗲𝗿𝘀𝗶𝘁𝘆 - 𝗧𝗮𝗿𝗴𝗲𝘁 𝟭𝟰: 𝗠𝗮𝗶𝗻𝘀𝘁𝗿𝗲𝗮𝗺𝗶𝗻𝗴 𝗼𝗳 𝗕𝗶𝗼𝗱𝗶𝘃𝗲𝗿𝘀𝗶𝘁𝘆 𝗩𝗮𝗹𝘂𝗲𝘀 - 𝗧𝗮𝗿𝗴𝗲𝘁 𝟮𝟭: 𝗞𝗻𝗼𝘄𝗹𝗲𝗱𝗴𝗲 𝗦𝗵𝗮𝗿𝗶𝗻𝗴 Huge shout to Aravind Ravichandran and his team for keeping us updated and Congratulations on a successful first EO Summit 🌍

20

How is the demand for AI capabilities shaping the future of data center technology? In an interview with DIGITIMES Asia, Judy Lin shares insights from Ranovus, a Canadian company making waves with its silicon photonics chips. Their partnership with MediaTek aims to revolutionize AI data centers with a 6.4Tbps Co-Packaged Optics solution. The collaboration notably enhances the capacity and efficiency of AI applications, offering a glimpse into the technological strides being made towards accommodating the exponential growth in compute capacity required by generative AI. This advancement is not just a testament to innovation but also highlights the critical role of optical interconnects in meeting the steep demand for efficient and powerful compute servers. https://lnkd.in/gcp8ikRD #AI #ArtificialIntelligence #Technology

5

This video highlights my ongoing MASc Thesis at U of T in the department of ECE supervised by Prof. Stephen Brown and Prof. Kevin Truong. Big thank you to Aman B. a PhD candidate at Caltech CNS in the Thomson group, for editing this video, inspiring many initial ideas for the project, and for helping through several breakthroughs this past week! Aman is a fellow EngSci grad from U of T! Also thank you to the Society for AGI for some very helpful discussions! A big inspiration for this project was the paper “Growing Neural Cellular Automata” (https://lnkd.in/gVcfejvn). The authors put a twist on Von Neumann cellular automata by demonstrating the interaction rules could be learned using neural networks to self-organize into desired 2D pixel patterns. Stephen Wolfram had already shown very simple cellular automata to be Turing Complete, so it was only a matter of time before their full possibilities started being explored. Moving from 2D to 3D, and fixed grids to free-moving cells required the insights of geometric deep learning, beautifully outlined in this proto-book: https://lnkd.in/gV4Yybk7. Geometric deep learning revolves around the idea of symmetry—which I think of as “the ways that something can be transformed without changing.” For example, a circle remains exactly the same when rotated in place. In a very real sense, that's what makes it a circle. In the more familiar case of bilateral symmetry, it's reflections over an axis that leave the object unchanged. Symmetry is the concept behind revolutions in modern mathematics (eg. group theory), physics (eg. gauge theory), and now deep learning (in architectures exploiting invariance and equivariance). CNNs led to Group CNNs (https://lnkd.in/g7AkqSK2), and eventually to the SE(3) Transformer (https://lnkd.in/gcV7MVzc). SE(3) is the special euclidean group, consisting of rigid translations and proper rotations in 3 dimensions—meaning this group inherently captures the structure of euclidean space. An architecture with SE(3) equivariance is indispensable for moving from fixed grids to freely moving cells. The power of Equivariant transformers can be seen through their use in AlphaFold, which outputs the 3D coordinates of all heavy atoms for a given protein (https://lnkd.in/gMSPc8t4...) AlphaFold recently won the Nobel Prize in Chemistry for its groundbreaking advance in protein structure prediction. The video of the frog developing was taken by Jan Van Ijken (https://janvanijken.com). Other inspirations include: Michael Elowitz, Rob Phillips, Karl Friston, John Conway, Carver Mead, Geoffrey Hinton, Richard Feynman, Eric Winfree, John Hopfield. I briefly foreshadowed a little bit of this project on the MLST Podcast a few months back (second half of that episode), and you can find that interview here (https://lnkd.in/gvXhKMKk)

112

10 Comments

Wow! CEO of NVIDIA, Jensen Huang, has a 60-person executive leadership team. He doesn't conduct 1:1 meetings or individual reviews for any of them. He provides on-the-spot feedback in front of the whole group. Many may see this as non-conventional and against best practice. I see an opportunity for healthy conflict, cohesive leadership, and clarity for the teams they lead. From experience, teams that model these 4 behaviors: -building trust -mastering conflict -achieving commitment -embracing accountability Always shine on any team, work or not. Would love to hear some thoughts...

7

2 Comments

Put the power of quantum computing right into your workflows. Access everything you need: hybrid solvers, SDK, open-source tools, libraries, code examples, demos, and comprehensive documentation. It’s all available through the Leap™ quantum cloud service. Begin building quantum applications with ease today: https://lnkd.in/eNJUJMx

10

1 Comment

Discover our global ecosystem of #NVIDIAInception startups. We support startups at every stage to innovate, build, and grow with special access to: 💡 Technical expertise ✨ Preferred pricing on hardware and software 🚀 Go-to-market support 🤝 Introductions to VCs and customers Take a closer look at how we support startups: NVIDIA for Startups

3

https://lnkd.in/gdnks2ap SK Group chairman met with Microsoft and OpenAI CEOs to reinforce global AI partnership network. #skhynix #ai #technology #innovation #business #Semiconductorindustry #electronics #semiconductor #supplychain #engineering #manufacturing #computerchips #semiconductors #chips #chipmaker #foundry credit: digitimes

7

✿核融合の先生　新着記事の紹介✿ _/_/_/_/_/_/ Commonwealth Fusion Systemsとは？核融合資金調達額No. 1企業 _/_/_/_/_/_/ あのMicrosoft創業者ビル・ゲイツ氏やGoogleが投資した、世界で最も大きな核融合スタートアップについて解説 #核融合　#フュージョンエネルギー #fusion Announcement of new article issued from The Nuclear Fusion Teacher. This article explains Commonwealth Fusion Systems, which is the biggest fusion start-up in the world. Reference: https://cfs.energy/ https://lnkd.in/gBQ_tJWq

2

If you are interested in LLM inference serving performance (aka tokenomics), there are a couple of threads on X from Yangqing Jia that are a must read.. He explains about various optimizations in inference serving including larger batch sizes, dynamic batching, quantization, speculative decoding, prompt caching etc.. He then digs into the factors behind the inference providers pricing for different models.. https://lnkd.in/grSYKjkQ Also a good related read would be Dylan Patel's post earlier in the year, where he did a detailed breakdown of the token economics from inference providers. https://lnkd.in/gfHkWbxF Definitely worth a read if you are looking at inference serving of LLMs. #llm #inference #ai #genai #artificialintelligence

71

1 Comment

Check out our new paper from Frederic H. and Ming Li at Atom Computing and academic collaborators: https://lnkd.in/g6NKQENX Neutral atoms have emerged as a leading platform for scalable, error-corrected quantum computing, and understanding the structure of Rydberg states is key to engineering high-fidelity interactions between qubits.

43

4 Comments

1/6 Today I attended the Google Developer Group - Vancouver #BuildWithAI conference. I joined the AI-Driven AR: Text-to-Image Magic, Vertex Search and Gemini in BQ, Building GenAI powered applications using Autonomous Agents, and Generative AI in Art & Design sessions. First impression: Excellent sessions from all the speakers and volunteers/organizers. I took pages of notes that I'll break down the lessons learned/key points, and some links for anyone who is interested! Unfortunately, I was not able to attend the rest of the sessions due to my cloning machine being out of order for the day. 🤔 Some main points/lessons I took from this event are: * Prompt engineering is very important and makes a significant difference in the responses you get from the AI tools. Ask bad ❓, get bad answers. I'll write more on this in Part 6. * AI will not steal your jobs. *Terms and Conditions apply. * There are a multitude of tools and ways to interact with AI, but not everything that is labeled AI does the same thing. 🤖 * Large Language Models (LLM) can be easily(ish) fine-tuned, but training them is a significant cost investment, so for most cases, the “custom” LLM’s or AI’s you see offered are just tuning the existing LLM’s models. The rest of the content learned will be spread out into 5 additional posts. There is too much information to share, too many links and too much inspiration for only 1 post. Part 2/6 – Session 1 - AI-Driven AR: Text-to-Image Magic Part 3/6 - Session 2 - Vertex Search and Gemini in BQ Part 4/6 – Session 3 - Building GenAI powered applications using Autonomous Agents Part 5/6 – Session 4 - Generative AI in Art & Design Part 6/6 – Final thoughts #AIWorkshops #MachineLearning #GoogleCloud #GenerativeAI #ContinuousLeaning #Tech #VertexAI #GeminiAI

7

1 Comment

https://lnkd.in/g3TJ8weV Nvidia, Qualcomm, AMD, others are considering raising prices due to the tight supply of TSMC 3nm manufacturing technology for their AI chips. 7 major chip designers in all (Intel, Apple, MediaTek, Google) are vying for TSMC’s 3nm capacity. The report says yield issues at Samsung have hampered its 3nm efforts, leaving customers with nowhere else to go but TSMC. Qualcomm will raise the price of its flagship Snapdragon 8 Gen 4 chips to over US$250 each due to higher production costs with TSMC’s 3nm process, citing supply chain rumors, compared to the $200 for Snapdragon 8 Gen 3 chips last year, which were made on 5nm process. 3nm wafers are said to be around 25% more expensive than 5nm. #tsmc #artificialintelligence #llms #generativeai #gpu #servers #datacenter #cloud #ai #Semiconductorindustry #electronics #semiconductor #technology #innovation #business #supplychain credit: ctee.tw

131

1 Comment

Recently, there has been a lot of discussion about TSMC's advanced packaging . Let us provide an in-depth introduction to the packaging of this foundry giant through their financial reports and the latest technology summit . Information shows that after Zhang Zhongmou returned to the company in 2011, he made up his mind to do advanced packaging . Yu Zhenhua, who joined the company in 1994, is the leader of this "secret" project of TSMC. CoWoS technology is TSMC's small test in this field. Their technology was first implemented on Xilinx FPGAs , and the InFO packaging derived from this has shined on Apple processors, and has since made TSMC's packaging famous all over the world. Popular science on TSMC’s advanced packaging technology According to Semiwiki, last year, TSMC merged their 2.5D and 3D packaging products into a single, comprehensive brand 3DFabric. Among them, 2.5D packaging technology CoWoS can be divided into CoWoS and InFO series. First look at CoWoS technology, which can be divided into the following types: 1. CoWoS -S Chip-on- wafer - on -substrate with silicon interposer for die-to-die redistribution layer (RDL) connections is celebrating its big break 10th year of series manufacturing. 2. CoWoS -R The CoWoS -R option replaces the (expensive) silicon interposer spanning the 2.5D die placement area with an organic substrate interposer. The trade-off with CoWoS-R is the smaller line spacing of the RDL interconnects—for example, 4 micron spacing on organic compared to the submicron spacing of CoWoS-S. 3. CoWoS -L Between Silicon -S and Organic-R interposer options, the TSMC CoWoS family includes a newer product with a “native” silicon bridge for (ultra-short distance) interconnect between adjacent die edges. These silicon wafers are embedded in an organic substrate and provide high-density USR connections (with tight L/S spacing) as well as interconnect and power distribution capabilities to (thick) wires and planes on the organic substrate. Note that CoWoS is designated as a "chip last" assembly process, with the chip attached to a fabricated interposer. Maximum package size and RDL enhancement CoWoS -S: 3X reticle (certified in 2021); CoWoS -R: 45X mask (3X in 2022), 4 RDL layers on organic substrate (W/S: 2um/2um), using SoC + 2 HBM 2 die stack for reliability certification; CoWoS -L: 1.5X reticle size reliability evaluation test tool with 4 local interconnect bridges between 1 SoC and 4 HBM 2 chip stacks; InFO _oS: 5X reticle (51mm x 42mm on 110mm x 110mm package ), 5 RDL layers (W/S: 2um/2um), currently under reliability evaluation; #business #india #china #europe #buy #globalbusiness #headhunter #usa #automotive #automotiveindustry #leadership #motivation #kindness #bestadvice #personaldevelopment #linkedin #hiring #careers #help #employment #happiness #jobs #health #global #world #jobchange #process #plasma #Knlowledgestudy #buyer #globalbusiness #Technology #recycle

3