WEBVTT 01:13:41.000 --> 01:13:50.000 opportunities, especially for the new type of accelerators, as you have seen in the presentation. The ideas that we have is leveraging 01:13:50.000 --> 01:13:59.000 potentially specific data sets and specific libraries, right? And obviously, MLIR will enable us to do this, this mapping. 01:13:59.000 --> 01:14:07.000 So, uh, that's… that's definitely a lot of opportunities there. There's a lot of discussion that we can have, and… 01:14:07.000 --> 01:14:09.000 Uh, they got it. 01:14:09.000 --> 01:14:13.000 Okay, right. 01:14:13.000 --> 01:14:14.000 Yeah. 01:14:14.000 --> 01:14:16.000 And Valerie, we're gonna have a longer, deeper follow-up conversation with the team on this, okay? 01:14:16.000 --> 01:14:20.000 Okay, that would be great. Thank you, but enjoyed the presentation. 01:14:20.000 --> 01:14:22.000 Very good. Um, Maya? 01:14:22.000 --> 01:14:26.000 So, I have a question that probably 01:14:26.000 --> 01:14:36.000 has a lot of different answers, but I know time is short, but uh… it seems as if when you're doing a chiplet and chip design, that the majority of the time is spent in verification. 01:14:36.000 --> 01:14:48.000 And I wondered, uh, what… how you were addressing that, or for the MVP concept, that isn't as important as getting the ideas out there. 01:14:48.000 --> 01:14:55.000 Yeah, um, you know, part of… part of our, uh, approach, I think, is to… is to try and. 01:14:55.000 --> 01:15:01.000 uh, what, maybe… sideline a lot of the verification. 01:15:01.000 --> 01:15:08.000 Uh, tasks, uh, through rapidly and… inexpensively. 01:15:08.000 --> 01:15:14.000 creating test hardware. So that our verification. 01:15:14.000 --> 01:15:30.000 of, uh, triplet packaging and triplet integration. comes from the use of testing, but maybe this is something Andreas could… Talk about how the definition of our chiplet interfaces. 01:15:30.000 --> 01:15:36.000 We'll also, uh, help address the verification challenge. 01:15:36.000 --> 01:15:43.000 Yeah, uh, hi, Maya. Uh, I think that there's a couple of things, uh… that, uh, that we're doing. One is, uh. 01:15:43.000 --> 01:15:56.000 The open source aspect of, uh… of some of the IP that we're developing, and so… You know, you certainly… there's less verification when you don't have the right code, so when you download. 01:15:56.000 --> 01:16:07.000 you know, TensorFlow, PyTorch, right? You… somebody else has done the verification, so… Yes, you do have to worry about how you compose those library functions and create the upper-level logic, but at least somebody has. 01:16:07.000 --> 01:16:15.000 unit testers and files and verify the lower-level stuff, right? So that's a net gain. And in hardware, we generally don't do that. 01:16:15.000 --> 01:16:22.000 the, you know, we do reuse some. IP out there, but for the most part, we write a lot of code. 01:16:22.000 --> 01:16:29.000 From scratch, and so we can say that verification, but I think the much more important thing is the fact that once you go into chiplets. 01:16:29.000 --> 01:16:43.000 you just short-circuit the verification altogether, especially all the physical stuff, the… running DRC on the deck, running signal integrity, power resolution. Once you have a die, you have a physical device that you can do a data sheet for. 01:16:43.000 --> 01:16:54.000 And then, uh, the verification goes down drastically, so… The idea of turning IP into a hardened physical device is sort of raising the level of abstraction. 01:16:54.000 --> 01:16:58.000 By a lot. Now we just have a box with pins on it. 01:16:58.000 --> 01:17:10.000 As opposed to, uh… a white box that you really have to verify every time you tape it out. And we see this a lot with, uh, especially with complex IP, like certies and memory controllers. 01:17:10.000 --> 01:17:17.000 Where even though we license it from big vendors, and we pay them millions of dollars, we still have to spend 6 to 9 months before we tape out. 01:17:17.000 --> 01:17:22.000 And imagine if instead of that, you had a memory controller chiplet that just worked. 01:17:22.000 --> 01:17:30.000 And all you had to do was hook up the pins properly, and you're done. You can imagine how much, uh, how much you can save in that, so… That's my… that's my spiel. 01:17:30.000 --> 01:17:31.000 Thank you. 01:17:31.000 --> 01:17:33.000 Thank you. Thank you, Andreas. 01:17:33.000 --> 01:17:42.000 So, let me just quickly say to Andreas, right now, at, uh, the OCP thing, ARM is presenting a chiplet 01:17:42.000 --> 01:17:44.000 their chiplet interface. 01:17:44.000 --> 01:17:49.000 Like, at this moment, it's on their screen, because Ron just sent it to me in a… 01:17:49.000 --> 01:17:53.000 I saw that. The more standards, the merrier, they say. 01:17:53.000 --> 01:18:00.000 Yes, I learned from Sadash Ankhor that we're in conflict with the. 01:18:00.000 --> 01:18:18.000 the OCP workshop today. Okay, um, I think that's all. Uh, any… any last questions or comments from anyone? 01:18:18.000 --> 01:18:28.000 Alright, if not, um… invitations open to all the Meerkat teams. Feel free to reach out to me, or any of our leadership team here at Decode. 01:18:28.000 --> 01:18:33.000 Um, we're happy to have follow-up discussions. 01:18:33.000 --> 01:18:34.000 Thank you, everyone. 01:18:34.000 --> 01:18:36.000 Thanks so much, Jim. Thanks, everybody. 01:18:36.000 --> 01:18:42.000 Thanks.