Shaoshan Liu

Reflections from the Silicon Horizon Last Thursday, Berkeley SkyDeck and Synopsys Inc brought together founders, investors, operators, and researchers for “The Silicon Horizon: The Next Decade for the Chip Industry.” It was energizing to help emcee a conversation that stepped back from the next node or the next funding round and focused on where this industry is really heading. What stood out most wasn’t any single prediction, but the consistency of the message across very different vantage points. We’re entering a decade where progress is less about brute-force scaling and more about how intelligently we design, integrate, and deploy silicon—from architecture and memory to packaging and system-level thinking, especially for AI-driven workloads. For startups in particular, speed to tape-out is becoming existential, not optional. And just as important, no one advances alone. The strongest insights reinforced how tightly coupled this ecosystem has become—academia, EDA and cloud platforms, foundries, and capital all moving together, increasingly organized around specific problems and workloads rather than general-purpose chips. It feels like a moment where collaboration and alignment matter more than sheer scale. Thank you to the speakers and panelists for sharing their perspectives so candidly: Borivoje Nikolic (UC Berkeley) Sridhar Panchapakesan (Synopsys) Krste Asanovic (SiFive) Ravi Subramanian (Synopsys) Vijay R. (Mayfield) Paul Rousseau (TSMC) And many thanks to Chon Tang for guiding a thoughtful panel discussion. It was great to see so many thought leaders join us and help make the event what it was. Looking forward to continuing these conversations as the silicon horizon comes into clearer focus. #SiliconHorizon #Semiconductors #ChipStartups #EDA #AIHardware #BerkeleySkyDeck #Synopsys

44

⭐ Excited to share our new blog post⭐ Intel Labs and Hugging Face Introduce enhanced Universal Assisted Generation (UAG) to 🤗 Transformers. Traditional speculative decoding requires the assistant and target models to share the same vocabulary. UAG mitigates this limitation by allowing the use of ANY target/assistant model pair, regardless of vocabulary differences 🤯. The latest UAG upgrade in 🤗 Transformers v4.50 now supports both greedy decoding and probabilistic decoding (sampling) for greater flexibility. See link to the blog post here: https://lnkd.in/dMBvcBk3 Thanks to the awesome collaborators: Jonathan Mamou,  Daniel Korat, Moshe Berchansky, João Gante, Lewis Tunstall, Nadav Timor, Moshe Wasserblat Roy Schwartz

80

1 Comment

Proud to have signed an MOU with the US Department of Energy this week. Cerebras has worked with DOE and its National Labs for years on hard science and national security problems. This agreement is about doing more of that - faster.   The goal is straightforward: use AI and AI+HPC to speed up scientific discovery and improve US R&D productivity.   This work supports the White House’s Genesis Mission... an effort to rethink how science is done, with AI as part of the core workflow.   Under the MOU, we’ll work together on things that matter:   ��� Large scientific data sets 👍 New AI and AI+HPC architectures 👍 Power, packaging, cooling, memory, and I/O 👍 Software and developer ecosystems 👍 AI "co-scientists"   The U.S. doesn’t lead in science by accident. It built world-class computing infrastructure and put it in the hands of world-class researchers.   That remains our playbook.

399

12 Comments

* NEW PAPER! * I'm ending the year on an incredibly important note, by sharing research highlighting the talent and insight of audio description (AD) professionals. These are the people who write, narrate, quality check, and produce narrations of important visual information shown in videos. These narrations are then mixed into videos, making the media accessible to blind and low vision (BLV) viewers. AI and other changes in working norms are impacting creatives. In particular, human-powered accessibility provisions such as AD are prime targets for automation. After all, there are a lot of inaccessible videos out there, and they should be accessible. We interviewed AD professionals including several who themselves are BLV, to learn how they go about their work and their perspectives on changing norms with generative AI's precipitous uptake. BLV AD professionals in particular encounter inaccessible work environments; there is opportunity to vastly improve the tooling AD professionals use. However, while automation could support in certain ways, such as through providing initial visual descriptions of scenes to BLV professionals, AD professionals emphasized the need for continual human guidance to ensure descriptions are appropriate to the media and providing rich immersive and descriptive experiences for BLV viewers. Disclaimer: I have started to publish non-anonymous work to credit creatives. We chose to anonymize AD professionals in this case given some of the results we share about inaccessible tools, ableism, and other forms of prejudice AD professionals have encountered. I had the pleasure of collaborating with Shaun Kane, Renee Shelby, PhD, Amy Pavel, Aidan Bryant, and Laura Eberly on this work. Read our paper here. https://lnkd.in/eyniCQ5M

169

11 Comments

Missy Cummings Keynote video from SafeComp 2025: "What California’s Self Driving Car Data Tells Us About Academia’s Shortcomings" Missy talks about the progression of types of reasoning with uncertainty, robotaxi incident reporting statistics, why the industry is getting statistics wrong, robotaxi crash types, sensor fusion, phantom braking, why Tesla fans are stalking Waymo robotaxis in Austin, the Cruise pedestrian dragging mishap, and a determinism/criticality heat map. "We have never needed safety research more than we need it right now." https://lnkd.in/efjCFBZM #SafeComp2025

78

6 Comments

At #OCPSummit25 in San Jose, I sat down with Ventiva CEO Carl Schlachte to explore a counterintuitive idea: lessons from highly constrained laptop thermal envelopes can help stabilize AI servers and racks. We cover hotspot behavior under mixed CPU/GPU loads, coexistence with direct-to-chip liquid, and why “inside-the-box” gains still matter as power density soars. Worth a read if you’re balancing acoustics, reliability, and throughput at scale. Check out the full article: Link in comments. #Ventiva Open Compute Project Foundation

10

1 Comment

Insights in robot training via robot videos from USC Viterbi (best paper at RSS 2025) plus Mapping America's Machine Shops, China's Policy Enabling Robotics, Mathworks Robotics Repository and Best Paper Awards and Finalists from CORL 2025 https://ow.ly/N9Az50XacU3

13

1 Comment

By now, I imagine many of you have seen Giacomo Indiveri’s very well written, and intentionally provocative, article in Neuron: “Neuromorphic is Dead. Long Live Neuromorphic”. As a neuroscientist in neuromorphic reading a paper targeting the neuroscience community, I wanted to respond to some of the points raised. I can appreciate many of Giacomo’s concerns and I agree with much of it. That said, some of the arguments raised in the paper are similar to those that initially kept me away from the neuromorphic community, a field that I have since found to be incredibly rewarding and (largely) welcoming. Specifically the idea neuromorphic must be analog. Neuroscience (and science in general) stumbles when people start worshiping the model system rather than the goal. When we fail to look at complementary systems, or actively dismiss them, everyone suffers. In the case of neuromorphic, analog CMOS is just a model system, much like C Elegans. Personally, I have gained far more intuition about the brain from large scale digital platforms like TrueNorth, Loihi, and SpiNNaker, which the article is dismissive of because they aren’t what Carver Mead envisioned 40 years ago. I don’t personally get much benefit from small analog circuits, because the neuroscience questions I have asked focus on computing questions rather than physics questions. Giacomo is fair to say digital platforms abstract away continuous time and neuron complexity, but unlike analog CMOS, digital enables extreme scales, brain-like circuit complexity, and programmability.  When I look at the brain, I see an elegant efficient parallel computing system, and the old school neuromorphs see resistors and capacitors. And that’s okay – worms, flies, mice, rats, and monkeys all fail to capture important aspects of human brains yet we still study them. It allows more people to ask more questions from different scales. Analog circuit design shouldn't be a prerequisite for neuromorphic than worm optogenetics should be a prerequisite for neuroscience. The digital neuromorphic technologies criticized in his paper have brought comp sci, applied math, and systems neuro into the neuromorphic field. There is no way that is a bad thing. There is a lot I wish will change. I have long wished the memristor and AI fields would take neuroscience more seriously. I have worked with both fields and they often opt for near-term gains rather than looking to the brain. But we aren’t going to convince people looking at a trillion dollar industry by saying you’re either with us or you’re against us. One of the things that we have strived for at the NICE Conference is a big tent approach to neuromorphic computing and brain inspiration. Analog or digital, silicon or memristors, stochastic or deterministic. It doesn’t matter so long as the goal is to advance our understanding of the brain and the brain’s ability to perform computations that still elude us. Neuromorphic computing was never dead. It just grew up.

220

30 Comments

𝗠𝗶𝘅𝘁𝘂𝗿𝗲 𝗼𝗳 𝗘𝘅𝗽𝗲𝗿𝘁𝘀 (𝗠𝗼𝗘): 𝗦𝗰𝗮𝗹𝗶𝗻𝗴 𝗟𝗟𝗠𝘀 𝗘𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝘁𝗹𝘆 𝘄𝗶𝘁𝗵 𝗦𝗽𝗮𝗿𝘀𝗲 𝗖𝗼𝗺𝗽𝘂𝘁𝗮𝘁𝗶𝗼𝗻 Large Language Models (LLMs) continue to grow in size, pushing the limits of AI capabilities but also introducing challenges in cost, memory, and inference speed. Mixture of Experts (MoE) offers an innovative approach by using sparse computation, activating only a subset of parameters per input. Let's explore recent advances in MoE architectures and how models like DeepSeek-v2 and DeepSeek-v3 are optimizing efficiency. 🔹 𝗖𝗵𝗮𝗹𝗹𝗲𝗻𝗴𝗲𝘀 𝗶𝗻 𝗠𝗼𝗘: 𝗥𝗼𝘂𝘁𝗶𝗻𝗴 𝗕𝗼𝘁𝘁𝗹𝗲𝗻𝗲𝗰𝗸𝘀 & 𝗣𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝗻𝗰𝗲 𝗧𝗿𝗮𝗱𝗲-𝗼𝗳𝗳𝘀 While MoE improves efficiency, it also faces key challenges: 𝗧𝗼𝗸𝗲𝗻 𝗗𝗿𝗼𝗽𝗽𝗶𝗻𝗴 𝗶𝗻 𝗟𝗼𝗻𝗴 𝗦𝗲𝗾𝘂𝗲𝗻𝗰𝗲𝘀: OpenMoE struggles with routing stability, sometimes losing tokens in long sequences. Fixed Routing in Pretraining: Early routing patterns can be inefficient post-training. 𝗗𝗼𝗺𝗮𝗶𝗻 𝗦𝗵𝗶𝗳𝘁 𝗜𝘀𝘀𝘂𝗲𝘀: MoE models may struggle to generalize across different data distributions. A recommended solution is incorporating instruction-following data in pretraining to enhance routing adaptability. 🚀 𝗗𝗲𝗲𝗽𝗦𝗲𝗲𝗸 𝗠𝗼𝗘: Smarter Scaling for AI Models The DeepSeek series addresses these issues with innovative optimizations: 🔸 𝗗𝗲𝗲𝗽𝗦𝗲𝗲𝗸-𝘃𝟮: 𝟮𝟯𝟲𝗕 𝗣𝗮𝗿𝗮𝗺𝗲𝘁𝗲𝗿𝘀, 𝟮𝟭𝗕 𝗔𝗰𝘁𝗶𝘃𝗲 1️⃣ Multi-Head Latent Attention (MLA): Cuts memory use by 93% with efficient KV cache storage. 2️⃣ Fine-Grained Expert Allocation: Balances shared and specialized experts across devices. 3️⃣ Device-Level Load Balancing Loss: Ensures even routing across devices, improving stability. 🔸 𝗗𝗲𝗲𝗽𝗦𝗲𝗲𝗸-𝘃𝟯: 𝗔 𝟲𝟳𝟭𝗕 𝗣𝗮𝗿𝗮𝗺𝗲𝘁𝗲𝗿 𝗠𝗼𝗱𝗲𝗹 𝘄𝗶𝘁𝗵 𝗡𝗲𝘄 𝗘𝗻𝗵𝗮𝗻𝗰𝗲𝗺𝗲𝗻𝘁𝘀 1️⃣ Multi-Token Prediction (MTP): Predicts multiple tokens at once for better efficiency. 2️⃣ Auxiliary-Loss-Free Load Balancing: Dynamically adjusts expert selection without added inefficiencies. 3️⃣ FP8 Mixed Precision Training: Reduces training costs significantly (~$5.6M for full training). 4️⃣ Extensive Post-Training: Includes context extension (128K tokens), SFT, RLHF, and knowledge distillation. 📊 𝗞𝗲𝘆 𝗧𝗮𝗸𝗲𝗮𝘄𝗮𝘆𝘀 ✅ Trained with 2.78M H800 GPU hours ✅ Performance rivals top closed-source LLMs ✅ Practical, scalable MoE for real-world deployment 🔮 𝗧𝗵𝗲 𝗙𝘂𝘁𝘂𝗿𝗲 𝗼𝗳 𝗠𝗼𝗘: 𝗘𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝘁 𝗔𝗜 𝗦𝗰𝗮𝗹𝗶𝗻𝗴 MoE is revolutionizing LLM training, making sparse computation viable at scale. While early MoE models had challenges, recent breakthroughs like MLA, MTP, and smarter load balancing are proving MoE's potential. DeepSeek-v3 shows that sparse models can match dense models, signaling a shift in AI scaling strategies. What’s your take on MoE architectures? Will they define the future of AI, or do dense models still have an edge? Let’s discuss! 👇 credit : Cameron R. Wolfe, Ph.D.

185

20 Comments

The most interesting paper I read last week, showing promising work on optical generative models that offload parts of the compute from electrons to photons. UCLA researchers demonstrated generating images by physical light propagation rather than digital computation. The approach maps diffusion model concepts to free-space optics: a shallow digital encoder converts random noise to phase patterns, then a trained diffractive surface processes light to generate images with the optical computation happening in under a nanosecond (overall speed is limited by the SLM refresh rate). This cross-pollination between fields (GenAI → Optics → GenAI) strengthens the entire research ecosystem. Moving from bits to light as the medium opens new possibilities for energy-efficient and significantly faster inference. Paper: https://lnkd.in/gC4By9Vv

39

1 Comment

Waymo’s June 2025 blog post explores how *scaling laws*—a concept familiar from large language models—apply to autonomous driving. Their research shows that increasing model size, training data, and compute power leads to "predictable improvements" in motion forecasting and planning, which are core capabilities for self-driving vehicles. Key findings include: - Motion forecasting quality improves following a power-law relationship with training compute. - Larger datasets and more inference compute enhance performance in complex driving scenarios. - Closed-loop simulation results (which mimic real-world driving) also scale predictably, confirming that better training inputs lead to safer, more capable AV behavior. Waymo’s study, based on 500,000 hours of driving data, suggests that autonomous vehicle systems can be systematically improved—just like LLMs—by scaling up intelligently. These insights could also benefit broader embodied AI fields like robotics. https://lnkd.in/gGx4xhxD

15

1 Comment