Artificial intelligence is often framed as a software story of bigger models, smarter algorithms, and faster training cycles. However, memory still has to deliver information on time. When it doesn’t, even the most advanced models stall. As an expert AI ASIC design engineer, Yarden Elnir works on the systems that determine whether next-generation AI can actually operate at scale.

The Physical Limits Holding AI Back

The AI boom has exposed a mismatch between workloads and hardware. Most large models still rely on general-purpose GPU designs initially made for graphics. While powerful, these chips face a bottleneck with data movement. Massive compute units sit idle while waiting for memory, which wastes energy and time. Engineers refer to this constraint as the memory wall.

This problem is not theoretical. It shows up as higher costs, slower inference, and strained infrastructure. For applications in healthcare, science, and large-scale automation, these inefficiencies matter.

Elnir approaches this challenge from a system-level perspective. He focuses on how logic, memory, and interconnectivity work together. His goal is to design hardware that keeps insight flowing rather than stalling.

Entry Into Apple Silicon Engineering

From an early age, Elnir developed a deep interest in mathematics, physics, and chemistry. That focus led him to one of Israel’s top institutions, Tel Aviv University. Once enrolled, he studied electrical and electronic engineering. Later, he contributed to an EU-funded research project on energy-efficient desalination technologies.

His transition into the industry happened fast. While still a student, Elnir was selected for a highly competitive role in Apple’s Chip Design Group. He later joined Apple full-time as an ASIC design engineer in the Apple Silicon Storage Group. In this role, he worked on the flash storage controller used across iPhone, iPad, and Mac devices.

The storage controller governs how data is accessed for nearly every operation on a device. Improvements at this level translate into faster boot times, smoother application launches, and more responsive systems for hundreds of millions of users. Elnir helped define microarchitectural components that improved performance and reliability across Apple’s entire product lineup.

The Dataflow Engine Inside Sohu

After several years at Apple, Elnir saw that the rapid growth of AI models was constrained by hardware that wasn’t optimized for today’s neural networks, prompting him to turn his attention to AI hardware. That insight led him to Etched, where he now serves as a member of technical staff focused on RTL design. The unicorn startup is valued at over $5 billion.

Elnir works on the Etched Sohu chip. It’s a domain-specific ASIC built for generative AI. The central data-flow logic feeds the matrix multiplication units. This logic determines how efficiently the chip can execute large models without stalling.

By designing out-of-order data fetching and predictive control mechanisms, Elnir helps hide memory latency and keep computation units active. The result is performance and efficiency that general-purpose hardware cannot reach.

A Trusted Authority in Industry Circles

Elnir is recognized as an expert peer reviewer for Information Fusion, a top-tier computer science journal. He has served as a judge for elite engineering company competitions, including MIT HackMIT and Stanford’s FAF Multimodal Hackathon. He was also selected for the Extraordinary Fellowship. It’s an honor awarded to fewer than two percent of applicants worldwide.

Within organizations, he serves as a mentor. In this role, he helps raise technical standards and evaluate new semiconductor engineering talent. His approach reflects a core belief that hardware cannot be patched later. Every decision must be precise.

Long-Term Access to Advanced AI

Yarden Elnir’s motivation is not speed for its own sake. He sees efficient hardware as a prerequisite for responsible AI adoption. Intelligence that wastes power and time cannot be scaled sustainably.

“I am a firm believer in the power of hardware and technology to transform our world for the better,” he says. By redesigning the physical infrastructure underlying AI, he aims to make advanced models more accessible, efficient, and reliable.