Validation. Power in the deploy phase of the three big phases, actions and deliverables, post silicon validation work, does the ASIC work? Does the product work? You create a list of bugs, you might have found bugs while you were simulating, creates an errata. It's not uncommon for CPUs to have from Intel and AMD. They publish or add over further CPUs. These are the known problems we have. These things are incredibly complicated. Chips can get extremely complicated. Many of these aren't show-stoppers but it's good to know that there is some errors, some bugs. Usually, not too much RTL simulation happens here unless you find a bug. It's important to understand what the root cause is so you may go back and try to recreate the bug that you found during the validation step and the real product by running RTL simulation. It's a relatively small number of test cases that you can run per hour of real-time because RTL simulation is the slowest simulation there is in comparison to. There's products for mentor called Veloce, cadence makes a Palladium. You can run a modest number of test cases here in one hour. Generally, you can think these as the big boxes, the Palladium ones have this pink cooling fluid. I missed it when I wasn't there the day of the Palladium got installed at the Seagate but the software guys were there and it seemed wrong. It seems like the hardware guy should have been in there but he said, "Oh, you missed it when they poured all the pink cooling fluid," and I'm like, "The cooling fluid was pink?" He said, "Yeah, it was bright pink." So, okay. They're big supercomputer machines that you can load your design into your ASIC into, and run at about one to two megahertz. So, that is way faster than our TL simulation, which is much less than that. I don't know the exact number is right now and it depends on the size of the design and how much memory that server has, and blah, blah, blah, but it's way faster. They have a huge amount of DRAM in there and every node in that design can be saved for a million or two or three million clocks back in time so you can simulate up to the point of the failure that you found in your ASIC, and you have this great trace, and you can load it into a waveform viewer and see where it went wrong. Then, another thing you can do is build FPGAs of your chip of your ASIC, and you can run the largest number of test cases in this scenario trying to figure out to get to the root cause of bugs in your ASIC, and you can run a large number of test cases here because you can gets these modern FPGAs. You can get them to run in the 50 to 200 megahertz range. So, the RTL's slowest, we got about one to two megahertz here and anywhere from 50 to 200 megahertz depending on the depth of the logic cone in your design and the nature of your design. You do give up some observability. The Altera and Xilinx have a way of ahead of time. That's Xilinx called ChipScope and you tell it what nodes inside the design you want to look at and they get routed out, depends on the FPGA, and you can route those over to a header, you can put a logic analyzer on there and capture that data to see what's going on. So, either ways you can figure out if you discover a bug, how to get to the bottom of it. Product integration. So, some companies call them engineering models. Sometimes, they're known as prototypes and build some prototypes, build customer tests units. Some companies call them qualification units. You want to get them out once you believe your product to stable enough and you're willing to risk, showing it to a customer. So it might not be completely done yet, the software firm or might not be completely written yet and completely verified and validated yet. But eventually, you get to some point and you say, "Okay we're passed enough tests where we can getsome call samples too," to customers because you want their feedback, and want their feedback because you want their sales. So, that's something that happens during this validation phase. Then, something a lot of electrical engineering and programmers don't think about is, are all these physical effects due to thermal shock and vibration testing? I get some more to saying that coming up here in a moment, and then validate your security. Do an internal audit again. You know they've done it before, just go back and revisit everything again. I think I mentioned this already. Hire an outside security form to attempt penetration testing of your product because these companies are really good at thinking sideways. So, I imagine a whole bunch of [inaudible] coming after your product, trying to break it. They're going to throw everything into book at your product and try and compromise it, and the power and glitching inside generally side-channel attacks and compromises, and measuring EM radiation, trying to determine keys, and whatever else that they're trying to extract. So, they'll figure out ways to compromise your security. So, this takes weeks and weeks. This doesn't happen in one or two weeks. But eventually, you get to the end of the validation. We're good to go and customers like, "When can you ship us your product? When can you ship us your product?" You say, "Okay. I'm going to start chipping and ramping volume next week," and then you enter production. You enter customer support phase at that point in time. So, validation. What does it mean? You build informed factor products. So, an informed factor product is can we use this? The clicker transmitter and receiver, plastic case, USB cable, gun antenna. This is in-form factor because this is the finished product. An outer form factor wouldn't have the case, it'd be a bear circuit board, it might be a circuit board that's maybe four times the size of the circuit board that's inside here, and it's got test probe points all over it so you can hook up scopes and logic analyzers to it to see what's going on. That's an outer form factor product. It's functionally equivalent, has an antenna, has got a USB connection on it, melts as it got, there's AC power there, and USB port put flash stick in, what they do with that. So, that's the difference between an in-formed factor in outer form factor. An outer form factor drives are useful because you can hook all this gear up to analyze what's going on, helps you figure out if you've got a problem or not when something goes wrong. But validation, when your in-formed factor product, you can put your product through a grueling series of tests. So, you'll run functional tests to prove, validate that all features and requirements are met, and you may uncover bugs during this. There's this notion of burn-in, you run some number of devices for some number of hours. I think at Seagate, we ran probably 500 drives for maybe 500 hours or something in that order. What you're looking for is things that fail quickly. Trace on a printed circuit board, it wasn't wide enough and it was vaporized, for instance by the current. If you got a batch a week components from some silicon vendors, some discrete component on there and one of those fails, a whole bunch of them fail in this batch of say 500, or here's my example, 1,000 devices for one 1,000 hours. So, you're looking for the things that fail quickly during burn-in testing. You put a thousand devices in, run them for a thousand hours, and nothing fails, you've got high confidence, but you don't have a weak link and easily failable point in your system. Some companies call this reliability demonstration testing. I use that term at the CA micron, RTD. Run shock vibration, and thermal stress tests: This is huge and this is an aspect that software people and electrical engineers often overlook and I've got some interesting examples here. Focusing on thermal here, typical CMOS silicon operating junction temperatures are in the range of 0 to 95C. That's what generally we want to operate CMOS. Your manufacturer of your chip may tell you you can go up to 125C and may even tell you, you can take it all the way up to 125C. I don't know anybody that allows their junction temperature to go up to 125C because bad things start to happen at that point in time. So, managing the thermal aspects of your product is very important.
