#1287 1987 · TSMC (Morris Chang) · Semiconductor manufacturing

TSMC promised never to design chips so any designer could trust it to build theirs

the problem

Small chip-design firms had nowhere neutral to manufacture, since fab owners like Intel also designed competing chips

background

By the mid-1980s a wave of small, capital-light semiconductor design firms wanted to compete with giants like Intel, but chip fabrication required fabs costing hundreds of millions of dollars that only large, vertically integrated manufacturers could afford. Those manufacturers made and sold their own chip lines, so a design startup that brought its blueprint to one of them was handing its intellectual property to a company that might well decide to compete with it directly, or simply bump the smaller client's production run behind its own higher-priority work. Morris Chang, a semiconductor veteran who had run Texas Instruments' worldwide semiconductor operations, recognized that this conflict of interest was strangling an entire potential industry of fabless design companies before they could start.

Taiwan, where Chang had been recruited to build a national technology strategy, had real strengths in disciplined, high-precision manufacturing but comparatively little presence in chip design or marketing — exactly the reverse of what a company trying to out-innovate Intel would need, but exactly the profile needed for a manufacturer that promised never to compete with its own customers.

what everyone would do

The available strategy for a new manufacturer entering semiconductors was to compete as a vertically integrated player like Intel or IBM — designing, marketing and selling its own chip lines alongside manufacturing — the exact model that was already blocking small design firms from finding a trustworthy fab partner.

what they saw

The scarce resource was not fab capacity, it was a manufacturer with no incentive to steal a client's design. A charter forbidding TSMC from designing chips made that promise credible in a way no rival could match.

the move

Chang founded Taiwan Semiconductor Manufacturing Company in February 1987 as the world's first 'pure-play' foundry: a company that manufactures chips to other companies' designs and by charter never designs, markets, or sells chips of its own, removing the conflict of interest that made every existing manufacturer a potential competitor to the design firms that needed it.

why it works

The mechanism works because it removes a conflict of interest that no amount of trust-building or contractual promise could otherwise fix — a vertically integrated competitor can always claim it won't misuse a client's design, but only a company barred by charter from designing chips at all can make that promise unconditionally credible. That credibility let an entire generation of fabless firms outsource manufacturing entirely, specializing in design the way TSMC specialized in fabrication, and the resulting division of labor let both sides go deeper into their specialty than an integrated competitor covering both could.

the payoff

Fabless firms like Nvidia and Qualcomm scaled on TSMC's fabs; by the early 2010s it held about 90% of global foundry operating profit.

where it breaks

The model only works where design and manufacturing can genuinely be separated without losing critical coordination — for products where tight, iterative co-design between chip architecture and process technology matters most, integration can still win, which is part of why Intel and Samsung kept both functions in-house. It also requires the foundry to actually maintain neutrality once scaled; if a pure-play foundry ever moved into designing its own competing chips, it would destroy the trust the entire model depends on overnight.

what came after

The design/manufacturing split TSMC pioneered restructured the entire semiconductor industry into fabless designers and pure-play foundries, a division of labor that now underlies the business model of nearly every major chip company that doesn't own its own fabrication plant.

references

[1]Morris Chang: Foundry FatherIEEE Spectrum, 2011spectrum.ieee.org