IMEC’s move into DRAM, Flash research marks shifting sands of semiconductor R/D

The Inter-University Microelectronics Consortium IMEC, based in Leuven, Belgium, announced at its annual research review meeting last week that the organization is expanding an on-going research program in front-end-of-line process technology at the 32nm half-node to include basic materials and devices for DRAM and non-volatile semiconductor memories (see here.) The move is a major departure for IMEC’s process research efforts, which in the past have focused almost exclusively on logic and SoC processes. As such, it tells a fascinating tale about the semiconductor industry as a whole.

From a purely scientific point of view, the move makes a great deal of sense. Never have there been so many materials, device design, and device fabrication questions open at the same time in both the DRAM and NVM spaces. Just about everything is up in the air. But IMEC’s decision goes deeper than scientific curiosity.

An explanation begins with a quick description of the consortium’s business model. IMEC is partially funded by the State of Flanders, one of the two principle pieces of Belgium. But the majority of its revenue comes from something like a subscription model. IMEC researchers survey the intermediate-term future of the industry and propose research topics that appear vital to the semiconductor players. These topics have to meet a couple of criteria. First, they must be what IMEC calls “pre-competitive”: that is, the topic must be fundamental enough that companies can cooperate on the research now, and then use the results to develop products which they will take out into the market to complete against each other later. Second, they must be significant enough that most of the major players in the industry will hesitate to take on the research by themselves.

In the past, such pre-competitive research happened mostly at the great private research foundations, such as Bell Labs and Watson Labs, and in universities. But those are gone. In more recent times, this research was done in ad-hoc partnerships between IDMs, and then the results were passed on to foundries. In this period, IMEC served as a vital link, a neutral ground in which competitive IDMs could cooperate with each other and with university researchers, achieve meaningful results under independent direction, and take away valuable learning.

Now, IMEC executive VP and COO Luc Van den hove says, even that model is breaking down. “The traditional research resources in the industry are drying up,” Van den hove said last week. “And we are seeing that foundries can no longer wait for IDMs to do the process development and pass the results along to them.” With more and more IDMs moving to a fab-lite—which in practice means almost research-free—business model, there are fewer and fewer IDMs and foundries to share in the enormous cost of pre-competitive research in the semiconductor space.

That appears to be the driving factor behind IMEC’s move to include DRAM and NVM targets in their most recent definition of process research. The logic industry has become so centralized that it can no longer support its own research through consortia. In fact, other observers have pointed out, the baton has passed from the IDMs to the equipment companies, working directly with the foundries on process development. But even the equipment companies, close as they are to process and materials sciences, are in need of shared resources.

This probably explains why IMEC has invited the top five DRAM vendors in the world to join in its front-end-of-line research project, along with major IDMs, equipment companies and a few fabless semiconductor houses. It’s a simple matter of spreading the economic burden—or if you will, taking a valuable product to a wider market—in an increasingly centralized world in which the risks of development are falling disproportionately on segments of the industry without the resources to fund them.

As research costs continue to grow, and as shrinking geometries take pre-competitive research deeper and deeper into the realm of pure science, one suspects that IMEC will have to reach into other pockets as well. In the end-game, only governments, through their economic development funds and their sovereign wealth funds (SWFs), can provide the kind of money necessary to continue advancing to finer geometries. But what a partnership involving a SWF or a hedge fund would look like is a puzzle—a puzzle worthy in itself of a research project.