> 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.
Suppose 2 reasonable sounding things.
1) There is no point running a process if you have a 99% failure rate.
2) Better (bigger chip, smaller transistors) chips are harder to make.
Consider the probability of failure per transistor, as something that drops exponentially via learning by doing.
Now it all makes sense. As you gain learning by doing experience, your transistor production becomes more reliable. This enables you to make bigger chips. Chip size is adjusted to get it within the optimal 10% to 90% failure window.
The more learning by doing there is in production, the more ambitious the chip designers can be.
The results you cite are counter-intuitive enough that I can't accept them at face value without some confirmation on how learning effects are supposed to spill over equally regardless of geographical location and the amount of communication across firms. Maybe the spillover comes from reading journal articles, people switching jobs between companies, reverse engineering products, and so on, but basically I don't buy it without further evidence. Extraordinary results require extraordinary evidence.
> 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.
Suppose 2 reasonable sounding things.
1) There is no point running a process if you have a 99% failure rate.
2) Better (bigger chip, smaller transistors) chips are harder to make.
Consider the probability of failure per transistor, as something that drops exponentially via learning by doing.
Now it all makes sense. As you gain learning by doing experience, your transistor production becomes more reliable. This enables you to make bigger chips. Chip size is adjusted to get it within the optimal 10% to 90% failure window.
The more learning by doing there is in production, the more ambitious the chip designers can be.
The 1994 link is wrong, I think you want http://klenow.com/LBD_Spillovers.pdf
lmao @nicholasdecker might want to report this
The results you cite are counter-intuitive enough that I can't accept them at face value without some confirmation on how learning effects are supposed to spill over equally regardless of geographical location and the amount of communication across firms. Maybe the spillover comes from reading journal articles, people switching jobs between companies, reverse engineering products, and so on, but basically I don't buy it without further evidence. Extraordinary results require extraordinary evidence.