Learning-by-doing in the Semiconductor Industry

Is there an economic justification for industrial policy?

The United States has been pursuing various policies to support semiconductor manufacturing in the United States. The CHIPS Act in 2022 set aside billions in tax incentives and direct subsidies for American manufacturing sites, and before that, we’ve long had Department of Defense involvement in subsidizing the research and manufacture of semiconductors. (See, for example, Irwin and Klenow (1995))

These policies are justified mainly on the grounds of national defense. There can, nevertheless, be an economic justification for it. If there is learning-by-doing which spills over across firms in the same country, but not between countries, and the gains from learning by doing are long-lived, then protecting firms from competition can improve outcomes for the country doing it. It is an empirical regularity that firms learn how to produce things more efficiently as they get larger, and that efficiency increases as market size increases. There are two reasons for this. First, firms discover technical information about how to make products which allows them to improve their efficiency. Second, producing more units means that there is a thicker market of people and companies working in the field, and you can better matches. Often the technical discoveries can spread only because of the spread of employees, so the two can tie in together. If the first reason predominates, then there is no reason to subsidize domestic manufacturing. If it is the latter, then an argument is possible.

Douglas Irwin and Pete Klenow have a paper from 1994 assessing where spillovers come from. They found that there were very strong learning curves within firms — marginal costs would fall at least 20% for every doubling in production — but that it hardly mattered where the production was occurring for spillovers between firms. A technical advance made by a firm in Japan will have larger effect for that firm than for others, but it will improve chips in America just as much as it will improve chips made by other firms in Japan.

In any event, the gains from learning-by-doing were specific to the batch. They took the form of having far fewer rejected chips from a particular design of semiconductor. At the beginning of the production run, perhaps 90% of chips would be defective; by the end, less than 10% would be. When a firm would start work on a new run of chips, though, they would go back to losing 90% of chips again. Industrial policy would not lead to a permanent advantage, but to a small gain at best.

More recent papers concur with the basic shape of Irwin and Klenow’s findings. Goldberg, Juhasz, Lane, Lo Forte, and Thurk (2024) estimate learning-by-doing from a proprietary data set, and show similar cross-country spillovers. Their estimates are considerably lower than Irwin and Klenow’s, though. Including cross-border spillovers gives a cost reduction of 8% for every doubling in production. Chip designers spread innovations from one firm to other manufacturers of their chips.

In short, industrial policy in semiconductors must be narrowly based on reasons of national defense. There is no sound economic argument whatsoever. It is striking how nonsensical US policy has been. In 1986, concerned that Japanese imports were too cheap, we pushed for the US-Japan Semiconductor Trade Agreement, which restricted how much Japan could export to the US, and required that they import chips from us; and then a few years later, sent executives at Korean multinationals to prison for anti-trust violations which kept prices high. We hadn’t made up our mind whether we wanted chips to be cheap or expensive, I suppose. Then with the CHIPS Act, rather than acting like we are in a serious struggle, we attached inane riders and have delayed implementation. That isn’t to say that there’s no reason to be concerned with where production is occurring. Fully 60% of semiconductor production is in Taiwan. We should simply not cloak our aims in the rhetoric of promoting American manufacturing and good paying jobs.