Best Place To BuildSharan Srinivas | CTO, Mindgrove Technologies | “You have to be crazy to want to make silicon”| Ep.9
At a glance
WHAT IT’S REALLY ABOUT
Mindgrove CTO explains building chips, India’s mission, and failure.
- Mindgrove raised about $8M in Series A, with returning investors doubling down and new participation from Rocketship VC and Mela Ventures, plus approval under India’s Design Linked Incentive (DLI) scheme worth ~₹15 crore upon milestones.
- The episode breaks down the semiconductor value chain—fabless design, foundries, and OSAT/ATMP packaging/testing—using concrete examples like TSMC (manufacturing), ASML (lithography machines), and the ‘die vs package’ distinction.
- Srinivas explains why ‘nm’ is now largely a node label, why leading-edge (e.g., 3nm) improves performance-per-watt but is exponentially costlier, and why 28nm is a practical “Goldilocks node” for embedded and industrial use cases.
- Mindgrove’s first commercial chip, SecureIoT (28nm), is positioned as India’s first commercial-grade high-performance microcontroller SoC built around the IIT Madras Shakti C-Class processor, bridging academic demonstration to sellable product.
- The discussion frames India’s semiconductor push as both strategic sovereignty (“new oil”) and employment/economic ecosystem building, emphasizing that fabs seed wide supplier and services job creation rather than being the sole job engine.
IDEAS WORTH REMEMBERING
5 ideasFabless startups win by focusing on design, outsourcing manufacturing complexity.
Mindgrove positions itself as a pure-play design house: it designs the chip, uses external foundries for fabrication and external OSAT/ATMP partners for assembly/testing, then sells packaged chips—mirroring models like Qualcomm and NVIDIA.
‘Nanometer’ is no longer a literal transistor dimension—treat it as a capability/cost tier.
Srinivas notes that 3nm is partly branding and no longer maps cleanly to a single physical dimension, but it still signals higher density and better performance-per-watt—at dramatically higher tool and manufacturing costs.
28nm remains highly competitive for embedded markets where constraints differ from smartphones.
Embedded/industrial systems often prioritize voltage ranges, electrical characteristics, reliability, and cost over extreme compute density; Mindgrove calls 28nm a “Goldilocks node” that balances performance and power without leading-edge expense.
Commercial-grade silicon is fundamentally different from a research demonstrator.
RISE/Shakti proved feasibility with academic tooling and tape-outs, but commercialization requires sellable-grade reliability, packaging/testing readiness, and use of commercial EDA licenses—turning “it works” into “it ships.”
SoC value comes from integrating the whole system, not just the CPU core.
SecureIoT is framed as a full “system on chip” including the processor, memories/caches, peripherals, and interconnect—similar in concept to Snapdragon, but far smaller and targeted to specific embedded use cases.
WORDS WORTH SAVING
5 quotes“You have to be a special grade of mental… to be able to want to make silicon in today’s world.”
— Sharan Srinivas
“It will come when you make it.”
— Sharan Srinivas (quoting Prof. V. Kamakoti’s response about getting a Shakti chip)
“The ecosystem is set up for you to fail safely. Are you willing to accept that failure?”
— Sharan Srinivas
“A three nanometer node is exponentially more expensive than a 28 nanometer.”
— Sharan Srinivas
“I’ll give you $100 million to spend. How long will it take you to spend? … Five minutes flat.”
— Sharan Srinivas
High quality AI-generated summary created from speaker-labeled transcript.
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome