You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
Have you ever wondered how SoCs, CPUs, and the core chips in all our high-tech products are created? How everything is adapted to fit equally well in laptops and gaming consoles? The importance of the etching process in chip manufacturing, what you need to know about it, and the misconceptions surrounding it? Or finally, the role of chip designers in driving innovation within our industry?
These are just some of the questions I had the opportunity to ask Stephen Robinson, the chief CPU architect at @Intel. A brilliant mind with 20 years of experience at the company, he was naturally able to answer all of these questions and shed light on the complexities of SoC design.
Summary: Interview with Stephen Robinson, Lead CPU Architect at Intel
In this in-depth interview, French YouTuber hosts Stephen Robinson, Intel’s Lead CPU Architect (recently promoted, with 20+ years at Intel, previously leading E-core development).
Key topics discussed:
Role & Background
Stephen manages Intel’s CPU architecture team: defining performance/power/area targets, microarchitecture, and ensuring the design can be delivered on schedule. He studied electrical/computer engineering at UT Austin (class of ~1996) after originally considering graphic design.
Building a CPU/SoC from Scratch
First decision: Instruction Set Architecture (x86 at Intel, vs ARM/RISC-V elsewhere).
Stephen personally loves the hands-on design/debug/iteration loop the most; deciding high-level IPC features under real-world constraints is the other rewarding part.
Intel’s Advantages & Legacy
Deep x86 expertise allows constant optimization that minimizes legacy overhead. Huge accumulated knowledge, tools, validation infrastructure, and experienced teams are massive barriers to entry for newcomers.
Modern SoC Challenges
Deciding what to integrate (GPU, NPU/AI accelerator, I/O, etc.) is critical; missing a trend (e.g., AI) can leave you years behind.
Advanced packaging (chiplets/tiles) is now essential because different blocks want different process nodes for performance vs cost.
An SoC today looks like “a motherboard shrunk onto a chip.”
Process Nodes
New nodes are enormously expensive and complex (EUV, gigantic machines, insane clean-room requirements). Benefits: smaller → lower power, higher frequency possible. Misconception: marketing “nm” numbers; reality is density, power efficiency, and design-rule complexity.
Goals & OEM Reality
Intel designs a flexible platform rather than a single end product. Must satisfy many OEMs with different priorities (gaming performance vs thin-and-light battery life). Same silicon behaves very differently depending on cooling, power limits, and firmware.
Development Timelines
~1–2 years of RTL design + ~1 year of closure/validation/power/timing → tape-out.
Another 1–1.5 years from tape-out to products in customers’ hands.
Intel runs 3–4 generations in parallel at different stages; architecture team must plan ~3+ years ahead while still reacting to late learnings.
Innovation vs Execution
Primary job is hitting targets on time; being late hurts the entire ecosystem. Innovation often means clever space/power savings that free up budget for new features, rather than revolutionary clean-sheet redesigns. Big “bet-the-company” startup designs frequently underestimate the full scope of real requirements.
Personal Highlight
Stephen’s favorite moment: when first silicon samples boot a full OS quickly with no major issues, the ultimate reward after years of work.
The interview closes with warm thanks (and a switch back to French), praising Intel for allowing such an open technical discussion with their lead CPU architect.
He said that clean-sheet designs can underestimate the scope and the full set of requirements. I’m not sure whether he was referring to AheadComputing.
