The semiconductor industry is often cited as a "strategic" industry in part because important learning-by-doing spillovers may justify special industrial policies. Documenting the precise nature of these spillovers is crucial for determining the advisability of such policies and is helpful for understanding the contribution of learning to endogenous growth. Yet existing empirical evidence on learning by doing in semiconductor production is scant and evidence on spillovers is nonexistent. Using quarterly, firm-level data on seven generations of dynamic random access memory (DRAM) semiconductors over 1974-92, we find that (a) learning rates average 20 percent, (b) firms learn three times more from an additional unit of their own cumulative production than from an additional unit of another firm's cumulative production, (c) learning spills over just as much between firms in different countries as between firms within a given country, (d) Japanese firms are indistinguishable from others in learning speed, and (e) intergenerational learning spillovers are weak, being marginally significant in only two of seven DRAM generations.

The semiconductor industry is often cited as a strategic industry in part because important learning spillovers may justify special industrial policies. Using quarterly, firm-level data on seven generations of dynamic random access memory semiconductors over 1974-92, the authors find that learning rates average 20 percent, firms learn three times more from an additional unit of their own cumulative production than from an additional unit of another firm's cumulative production, learning spills over just as much between firms in different countries as between firms within a given country, and intergenerational learning spillovers are weak.