When the transistor was first developed, there was no way to extrapolate the hard-to-make, unreliable component’s success trajectory. Early bipolar transistor manufacturing techniques involved high temperatures that caused dots of liquid indium to diffuse into the germanium transistor body using a “precise” amount of time to control diffusion depth. It was more akin to alchemy than chemistry and it didn’t even slightly resemble today’s precisely controlled lithographic and doping manufacturing processes. A lot had to happen before transistors could be mass produced and the techniques were perfected by several companies—some no longer in business. Even more had to happen before we could commercialize the concepts of an integrated circuit.
It seems to me that 3D chip stacking is in a similar situation today. We’re not precisely sure what 3D manufacturing techniques will become mainstream or which products will most benefit from 3D chip stacking. We can see a lot of technical problems such as thermal stress, test access, and reliability that need to be overcome and we can see some potential solutions—but we’re not sure which will work best.
A 3D Ecosystem discussion at the Global Semiconductor Alliance (GSA) Memory Conference covered these topics and more on March 31. The panel was moderated by Matt Nowak, senior director of engineering at Qualcomm and the panelists included Bill Chen, senior technical advisor, ASE Group (OSAT); Kyowon Jim, vice president, product planning, Hynix Semiconductor (memory); Paul Kempf, vice president, silicon, Research in Motion (customer); Suk Lee, director, design infrastructure marketing division, TSMC (foundry); and Dave Noice, fellow, Cadence.
Richard Goering has covered the discussion in detail in his blog “3D IC Ecosystem Panel: Different Views, Challenging Questions”. When I read the discussion, I saw that there are many potential reasons for using 3D chip stacking: bandwidth, product volume, and cost are just three. Yet it appears from the discussion that these three dimensions and the others are not closely linked. It’s possible to push 3D stacking in one direction at the expense of the others, depending on where you need to go.
What is clear is that a lot of companies are going to pick up this 3D ball and run with it. There may be simply be multiple goals so that not all of the players will be running in the same direction.