真正改变世界的技术，为什么一开始都不被看好？| S10E16

Summary

This [[WhatsNextKejiZaozhidao|What’s Next｜科技早知道]] episode has [[WangBo|汪波]], author of [[ABriefHistoryOfChips|《芯片简史》]], use semiconductor history to explain why important technologies often look weak at birth. It connects [[MooreLaw|Moore’s Law]], [[MOSFET|MOS transistors]], [[ElectronicDesignAutomation|EDA]], memory demand, Huawei’s Tau Law, and Disruptive Innovation to the AI era. The durable synthesis is that world-changing technologies often begin with worse yield, heat, speed, stability, or tooling, then win when a small scalable advantage compounds through industry coordination and practical know-how.

Key Claims

• Moore’s Law began in a commercial and strategic context: early integrated circuits were hard to sell against discrete transistor assembly, so Gordon Moore’s forecast also helped make a future roadmap legible.
• The episode treats Moore’s Law less as a physical law than as an industry-coordination target that helped companies align manufacturing, design, investment, and expectations.
• Early objections to integrated circuits were technically reasonable: yield, heat, and missing [[ElectronicDesignAutomation|EDA]] tools made dense integration look impractical.
• Tau Law is compared to early Moore’s Law as a possibly mobilizing metric rather than a proven natural law; its credibility still depends on tools, yield, power, cost, and shipped results.
• The post-Moore menu includes continued transistor scaling, Advanced Packaging, Semiconductor 3D Stacking, HBM-style memory proximity, chiplets, and more speculative “beyond Moore” materials.
• Bell Labs did not prioritize the [[MOSFET|MOS transistor]] because [[BipolarJunctionTransistor|BJT]] devices better fit [[ATT|AT&T]]’s immediate switching needs, while MOS looked slower and less stable.
• Fairchild Semiconductor and Intel-style iteration later made MOS valuable because its simpler, cheaper, more integrable structure mattered more as density scaled.
• The episode frames the MOS story as Disruptive Innovation: a new route can lose on incumbent metrics while quietly winning on a different scalable dimension.
• The Google transformer/search example and the Intel/Nvidia comparison make organization part of the technical story: existing cash flow and evaluation systems can make self-disruption harder than invention.
• AI can assist semiconductor manufacturing and chip design, but the source keeps new process knowledge, undefined tool problems, cross-team communication, and [[DomainKnowHowMoat|domain know-how]] on the human side of the boundary.
• Broad reading, history, philosophy, literature, and biography are presented as professional infrastructure because they build judgment, analogical thinking, and a richer human world model under AI pressure.

Key Quotes

“以时间换空间” - the episode’s shorthand for Tau Law as a delay-reduction route under process-node constraints.

“更多摩尔、扩展摩尔和超越摩尔” - the three post-Moore paths discussed for semiconductor progress.

“向内求取” - Wang Bo’s closing phrase for finding human positioning while using AI as a tool.

Connections

• What’s Next｜科技早知道, Wang Bo / 汪波, and A Brief History of Chips / 芯片简史 - show, guest, and book context.
• Moore’s Law, Gordon Moore, Electronic Design Automation, and Semiconductor Supply Chain - semiconductor-roadmap and toolchain context.
• Bell Labs, Muhammad Atalla, Dawon Kahng / 姜大元, MOSFET / MOS Transistor, Bipolar Junction Transistor, Fairchild Semiconductor, and Intel - historical MOS-versus-BJT adoption branch.
• Huawei, Tau Law, System-Level Semiconductor Optimization, Advanced Packaging, Semiconductor 3D Stacking, and Constraint Driven Engineering Strategy - post-Moore and China constraint-response branch.
• Disruptive Innovation, Non-Consensus Innovation, Large Company Organizational Inertia, Google, Nvidia, Jensen Huang, and Andrew Grove - organization and self-disruption branch.
• Human Judgment Under AI, Domain Know-How Moat, Domain Expert Alignment, and Battery Manufacturing Know-How - AI-era career and tacit-knowledge branch.

Contradictions

• No direct contradiction found. The source reinforces the existing 华为的「韬定律」，是创新还是噱头？｜ Bonus position that Tau Law is more defensible as a system-time metric than as a literal replacement for Moore’s Law.
• It adds a more historical and optimistic analogy: early Moore’s Law and MOSFET / MOS Transistor skepticism show why a weak-looking route can later matter, but the episode still leaves Tau Law dependent on EDA, yield, heat, power, cost, and product proof.
• Contemporary industrial examples, including Huawei, Nvidia, Google, and AI’s role in chip design, are source-reported podcast claims and were not independently verified during ingest.