Technology

LLMs on Analog In-Memory Computing Based Hardware (IBM Research, ETH Zurich)

Source

SemiEngineering

Published

Dec 1, 2025

TL;DR

AI Generated

A technical paper by IBM Research and ETH Zurich introduces a method to adapt large language models (LLMs) for execution on noisy, low-precision analog hardware, improving speed and power efficiency for neural network inference. The method enables high-capacity LLMs to achieve performance comparable to traditional architectures despite analog noise and quantization constraints. The paper also demonstrates benefits in test-time compute scaling and the adaptability of analog foundation models for inference on low-precision digital hardware. This work bridges the gap between LLMs and efficient analog hardware, offering energy-efficient foundation models.

Read Full Article

CEO Interview with Xianxin Guo of Lumai

Xianxin Guo, CEO of Lumai, discusses the company's optical computing technology for AI and data center acceleration, aiming to address power efficiency and scalability limitations of traditional silicon-based approaches. Lumai's hybrid optical-electronic design enhances compute efficiency by leveraging light for key operations, reducing energy consumption and breaking through AI system bottlenecks. The technology is well-suited for high-throughput AI inference workloads in data centers, offering a more cost-effective and scalable solution. By focusing on optical compute, Lumai differentiates itself from competitors and aims to redefine AI compute efficiency for long-term scalability and performance gains. The company engages with customers through collaborative discussions and partnership-driven approaches to integrate optical computing seamlessly into existing AI infrastructure.

SemiWiki•

5 days ago

Intel launches Wildcat Lake as Core Series 3 for value laptops and edge systems — six consumer SKUs built on 18A promise 'all-day' battery life

Intel has unveiled its Core Series 3 mobile processors, known as Wildcat Lake, featuring six consumer SKUs and one edge-only variant, all built on 18A technology promising extended battery life. The lineup includes various configurations with P-cores, E-cores, NPUs, and Xe3 integrated GPUs, offering up to 40 platform TOPS and hybrid AI capabilities. Memory support tops out at LPDDR5x-7467 or DDR5-6400 in a single-channel setup, with performance improvements compared to previous generations. Initial laptop designs from Acer, HP, MSI, and others are set to launch, targeting students, small businesses, and edge deployments.

Tom's Hardware•

2 weeks ago

Analyzing Elon Musk's TeraFab — A step towards Tesla and SpaceX's partial vertical integration, or an unattainable dream?

Elon Musk's TeraFab project aims to produce AI logic chips and HBM memory, requiring trillions of dollars and hundreds of fabs, posing challenges in capital, equipment availability, and workforce expertise. The project's scale exceeds current industry capacity, with constraints in lithography tool supply and process technology development. TeraFab would need hundreds of thousands of construction workers and over 300,000 skilled employees, raising doubts about its feasibility. While ambitious, the project may serve as a step towards partial vertical integration for Musk's companies rather than a complete transformation of the semiconductor market.

Tom's Hardware•

4 weeks ago

MIT Technology Review

Google DeepMind wants to know if chatbots are just virtue signaling

Google DeepMind is exploring the moral behavior of large language models (LLMs) to determine if their actions in roles like companions or therapists are trustworthy. While LLMs have shown moral competence, there are concerns about their reliability, as they can change responses based on feedback or formatting. The researchers propose rigorous tests to evaluate LLMs' moral reasoning, including challenging them with variations of moral problems. Additionally, they acknowledge the challenge of designing models that cater to diverse values and belief systems globally. Overall, understanding and advancing the moral competency of LLMs is seen as crucial for the progress of AI systems aligned with societal values.

MIT Technology Review•

3 months ago