For many of the most advanced electronics systems, overall system performance is affected by how well some of the more critical ICs interface with those they interact with. A very common example is how quickly a logic chip can retrieve data from a memory IC.
The response has been to completely reimagine how ICs are constructed. Instead of having each chip encased in its own package, the industry has devised a number of ways to combine two or more chips in a single package. Significant performance boosts have already been demonstrated with some of these new packaging technologies.
Novel packaging technologies have been around for a decade or more, but there are still many technological areas left to explore.
This week, CEA-Leti, a French research institute that specializes in micro and nano technologies, announced a collaboration to explore new packaging technologies with Intel. Don Scansen is an old hand at EE Times. He was away for a few years, but we’re quite pleased to have him contributing again. He wrote about the collaboration between Intel and Leti, so we gave him a buzz at his home in the Great White North to asked him about it.
Don, what I’d like to ask you first is to go over what Intel and Leti announced today.
DON SCANSEN: Leti actually announced an initiative with collaboration with Intel to develop their 3D packaging to enhance the high-performance computing space. Intel, as far as I know, hasn’t made any mention of the collaboration so far. But the Leti announcement is about 3D packaging for high-performance computing.
BRIAN SANTO: Excellent. What is it about this deal that’s significant? Does Leti have technology that Intel does not? Does Intel have technology that Leti does not? What would you expect, or what does Leti expect to get out of this collaboration?
DON SCANSEN: It’s hard to say what specific technology Intel may have that Leti has. I think what Intel brings is their product side and know-how in manufacturing and in ramping up volumes, and to be able to help Leti commercialize some research technology that they have. That will help them, I think even in their research phases, to learn some things about the volume manufacturing that may help them tailor what they’re doing a little bit more.
From Intel’s side, if you want, I think there’s something that we may not know exactly which combination of things that Leti’s bringing. What I understand is, there’s interposer technology that they’re bringing for integrating chiplets. And I would expect as well that Leti may have some specific techniques that Intel can benefit from in the chip stacking space, so basically the die to interposer bumping in this connections.
BRIAN SANTO: The packaging part of the technology is getting more and more interesting. I think AMD, when they just bought Xilinx, one of the things they were enthusiastic about was that Xilinx has some packaging technologies that they’re quite adept at, that they’ll bring to the table. Packaging is becoming a really important element of the technology, isn’t it?
DON SCANSEN: I think so. It’s something that colleagues of mine talked about many years ago, back in the old semiconductor insights days. Integration and from the very beginning of chip design, packaging had to be considered. And now we see that really becoming relevant with the whole idea of the chiplet. I think that the Intel collaboration with Leti here is trying to keep them ahead to a certain degree. The AMD purchase of Xilinx is actually significant. Intel wants to stay ahead in high-performance computing for the cloud.
We don’t exactly know what Leti may have to offer yet with all the work they’re doing. I know that (?) interposer for a very long time Leti has had optical interconnects for that on their roadmap. That may be something that might not be relevant today with the work that’s going on, but Intel may have their eye on that for the future.
BRIAN SANTO: Right. If we don’t know the specific technologies that might be involved, we do know that it’s generically about packaging, and we do know that the ultimate goal is to remain ahead in high-performance computing, where AMD and Xilinx are teaming up and where Nvidia and Arm are likely to be teaming up, right?
DON SCANSEN: Yeah, exactly. I think the parallels and the comparisons are easier to draw between Intel, because they were the first to acquire one of the top two FPGA makers in Altair, and now with AMD doing the same.
The FPGA is part of the AI accelerator idea that’s used in the cloud, and right now those are basically separate cards. So I think part of what this ever-expanding, which is sort of a long suggestion of where things are going there. The shrinking I guess of things into the package that we might see those FPGAs brought in somehow more into the same package of the processor for the acceleration. There’s all kinds of things once could speculate on what Leti might have that makes that more feasible.
BRIAN SANTO: It’ll be fascinating watching the data center market and the supercomputer market moving forward then.
Anything I haven’t asked about this relationship that is interesting or intriguing? Any questions that are left unanswered?
DON SCANSEN: I think there are a lot of questions left unanswered. But I don’t want to pose those to you right now because I definitely don’t have the answers… yet.
BRIAN SANTO: We’ll keep reading your blog, your op ed pieces for EE Times, then.
DON SCANSEN: I certainly hope so! I hope that I have something worthwhile there to offer you.
BRIAN SANTO: Great. Don, thank you very much for your time. I appreciate it.
DON SCANSEN: Thanks, Brian.
BRIAN SANTO: You can read Don’s story at eetimes.com. It’s called, “5G and Interconnect: Posing Actively.” To find more stories from Don, click on the navbar button that says “Perspectives.”