AMD's Zen 5 processors have a flaw in their RDSEED instruction that compromises cryptographic security by returning too many zeroes, risking private key exposure.
About:
Zach is the author of a tech blog on Substack, where he writes about building chips and other technology-related topics. His blog has hundreds of subscribers.
Website:
Specializations:
Incoming Links:
Subscribe to RSS:
The author argues that LLMs (Large Language Models) are inadequate for generating high-performance chip designs due to the unforgiving nature of the design process, which requires high-quality code in Verilog. While reinforcement ...
The blog post discusses the vulnerabilities of Trusted Execution Environments (TEEs) and Secure Enclaves (SEVs), particularly focusing on the Heracles attack on AMD's SEV technology. It explains how TEEs, while faster and more eff...
The blog post discusses the challenges and lessons learned in commercializing unconventional computing technologies, which differ from the traditional Von Neumann architecture. It covers various methods, including processing-in-me...
The blog post discusses the overlooked security vulnerabilities in GPU-powered systems, particularly focusing on a new attack method called GPUHammer, which exploits the rowhammer vulnerability in GPU memory. The author highlights...
Crossbar's open-source security chip design has critical flaws that expose it to side-channel attacks, highlighting the need for better collaboration in open-source hardware development.
Nvidia's $20B acquisition of Groq, now clarified as a licensing deal, raises questions about the valuation and strategic motives behind the move.
The article discusses Extropic's recent announcement of their XTR-0 devkit, addressing claims that it vindicates the company against accusations of being a scam. The author critiques the efficiency claims of Extropic's chip, compa...
Tenstorrent, an AI startup led by Jim Keller, has acquired Blue Cheetah, an analog design startup. The acquisition raises questions about Tenstorrent's business strategy, as most of its revenue comes from licensing RISC-V-based AI...
The blog post discusses the challenges faced by startups in the electronic design automation (EDA) industry, dominated by Cadence, Synopsys, and Mentor Graphics. Despite complaints from chip designers about the inefficiency and ou...
OpenAI partners with Cerebras for AI chips despite higher costs, as availability becomes crucial amid soaring demand and limited manufacturing capacity.
Zach's Tech Blog celebrates a successful year with reflections on personal growth, challenges in chip design, and the significance of secure AI hardware.
Irreducible, a company focused on using FPGAs to accelerate zero-knowledge proofs (ZKPs), shut down after pivoting to software implementations. The post discusses the challenges faced by ZKP hardware companies compared to successf...
5.9Why did Tesla Dojo fail?
2025-08-11 • machine learning tesla classdojo nvidia ai density ai
The post discusses Tesla's decision to shut down its Dojo supercomputer for training full self-driving models, attributing the move to both executive departures and technical challenges. It highlights the complexities and financia...
5.8Taalas is what Etched should have been.
2026-03-09 • technology chips nvidia ai taalas
Taalas's AI chips offer remarkable speed but face significant limitations and costs, making them better suited for consumer applications than enterprise use.
Cadence Design Systems has acquired ChipStack, a startup that uses AI to enhance silicon verification processes in chip design. This acquisition reflects a common trend in the EDA industry where startups are bought by larger compa...
0Why did Untether AI fail?
2025-06-16 • generative ai ai chips ai vision inference
Untether AI, an AI inference chip startup, is shutting down and their engineering team is joining AMD. The company focused on AI inference chips and software, but ran into issues due to the rise of large generative models. They mi...
SambaNova Systems, a well-funded AI chip startup, has pivoted away from their original goal of offering a unified architecture for both training and inference. They have laid off 15% of their workforce and are now focusing entirel...
The post discusses the delays in building new IP cores for various connectivity protocols, video codecs, and memory interfaces. It explains that the increasing complexity of standards and the need to maintain legacy IP cores are t...
The post discusses the development of a chip at Normal Computing that enables unprecedented scalability for thermodynamic computing, focusing on large-scale generative design, scientific computing, and probabilistic reasoning algo...
Softbank is acquiring companies to build an AI datacenter empire to rival Nvidia. They have acquired ARM, Graphcore, and Ampere to leverage their synergies. Nvidia's key strengths are the raw power of its GPUs and providing turnke...
Big tech companies like Amazon, Microsoft, and Google are competing to build practical quantum computers. They are using different approaches to address the challenge of noise and quantum decoherence. Google uses well-understood q...
The text discusses the potential of LLMs in designing chips, particularly in generating Verilog. It also explores the challenges faced by companies selling AI-powered chip design tools and the potential of using AI tools to build ...
Forbes reported that Meta is considering buying South Korean AI chip startup Furiosa AI. Furiosa's RNGD uses a flexible architecture that maximizes data re-use, but under-performs Nvidia on pure computational power. Furiosa's chip...