To fab or not to fab ...
1 min read
That is, indeed, the question which appears to be occupying Qualcomm's senior executives as they look to secure adequate supplies of the chips which power many of today's smartphones.
It's another example of the cyclical nature of the semiconductor industry. Whatever the issues, it seems the industry will swing violently from one extreme to the other in relatively short timescales. One of the latest swings involves manufacturing.
Over the last decade or so, the rise of the foundries has encouraged some semiconductor companies to adopt the so called 'fab lite' strategy; outsourcing their manufacturing requirements to some extent. But there is also the fabless sector – including chip developers such as Qualcomm – which relies entirely on foundries.
But fab lite – even fabless – only works while there is capacity in the foundry network. Once capacities become constrained, it then becomes a question of how good is your relationship with the foundry?
Things have come to a head with the move to the 28nm process node. On its own, the move to 28nm shouldn't have created the roadblock it has. What has complicated the issue is that many companies have decided to make 28nm chips while the foundries continue to bring the process up to speed. Even long standing customers have had to deal with the uncertainty.
It's not unreasonable to conclude that things may not get any better, so what to do if you're that chief executive relying on the foundry network to manufacture huge volumes of chips to satisfy your demanding customers? Shutting fabs is easy for fab lite companies; bringing them back online at the latest process node is a challenge. Some fab lite companies will be able to get more from their existing fabs; others won't have that luxury.
But what if you're Qualcomm? It's said that Qualcomm's manufacturing requirements equate to the output of one of TSMC's fabs. On that basis, might Qualcomm decide that it's better to be master of its own destiny?
But – and it's a very large 'but' – building and running your own fab is a big step. All of a sudden, you have to manage the peaks and troughs of demand, the process technology challenges, the yield issues and so on.
One company has mastered this. While foundries have to chop and change between devices, Intel can focus on optimising its manufacturing process for microprocessors.
Qualcomm can certainly afford the $5billion or so needed to build a fab. The question its board will be considering is whether it wants to make such a big step on to the white knuckle ride of semiconductor manufacturing.
