The second keynote on the first day of ARM TechCon was a double act with Greg Yeric, a fellow from ARM's research labs in Austin, followed by Mike Muller, ARM's CTO. Greg Yeric I have seen Greg talk before, and he seems always to have the misfortune to be addressing the audience at the wrong level. The first time I saw him talk was one of the plenary presentations at last year's IEDM. He talked a bit about devices but a lot about system issues like standard-cell architecture and routing. This went over the heads of most of the audience, I think, who were device people for whom a carefully constructed single transistor is a thing of beauty. They didn't want to hear that IC designers don't really care about typical performance since they have to design three (or more) standard deviations from that, so reducing variation is probably more important than the headline typical performance that the device people like to report. This time Greg was talking at TechCon, and while there were some chip guys there, most of the audience were people involved in software development of one form or another. This time Greg's discussion of future process optimizations and whether EUV will save the day seemed to fly under the radar. As a guy with silicon in my veins, I enjoyed both presentations. Greg knows a lot about process stuff but also a lot about designing SoCs, which is a rare mixture. Greg started out by pointing out some widely believed falsehoods: 28nm was not the best node ever We are continuing to scale costs However, pitches are not scaling but the improvements are coming, such as one-time fixes like reducing fins per transistor and reducing standard cell height. There is a lot of empty space in a standard cell largely because we can't put a contact to gate over the transistor. Greg used to be laughed for even suggesting the idea but no longer. What if we could do a 5nm technology with five track standard-cells with two fins per transistor? A self-aligned gate contact is key. This is no longer just a process development activity, but requires design technology co-optimization (DTCO). Costs are affected by improved equipment. Greg gave the example of steppers. These are now 50% faster than at 28nm. In fact, their performance is positively amazing. They expose a wafer in 15 seconds, with the chuck moving at 1M/s (developing 10G acceleration) and aligning to 2-3nm. This is equivalent to dropping England on the globe and aligning it to the precision of one housing brick, 5 times per second. For 7nm, it looks like we don't get EUV, but Greg is assuming we do for 5nm, or else he will have to be less optimistic. Most of the cost in a wafer is now in the back-end metal stack. There are severe parasitic loads. There are some tricks, such as air gaps and low-R metals such as cobalt and ruthenium, but that doesn't solve the problem completely. Beyond transistors, we need better memories. Another one-time trick is 3D with wafer-to-wafer accuracy good enough to allow standard cell partitioning. We will need new tools to exploit it but there are lots of possibilities such as putting p and n transistors on different wafers, or putting high speed on one wafer and low power on another. This will all take a lot of money. The R&D costs for 2/3nm will be unprecedented. There is lots of good materials research going on, but largely from people who don't understand what designers need (see my opening paragraph!). ARM is playing their part and have now joined imec, who have a 3nm pilot line including an EUV stepper. Greg is confident we will have a good 5nm node. For 3nm, he is not so sure. Mike Muller Next up was Mike Muller, ARM's CTO. He wondered whether you could trust your health care supplier. He talked about a lot of diseases, such as cancer and malaria, which he fortunately does not have. One thing he talked about was a competition that they set up for wearables. The entries had to be innovative, low cost, and reliable since they were targeting the world, not just rich countries. The entries were truly global, and they selected 10 finalists and eventually two winners. The first winner was a RFID tag for tracking vaccination of very young babies. It only has to be used for two to three years. As the child grows up, it is recycled. The idea is to close the "last mile" gap on patient records in underdeveloped countries, where offices may have full access to records, but in the field not much. Mike then got to what he admitted was his least favorite part of the presentation: new product announcement. Security is obviously a big issue with the internet of things (IoT) in general and medical devices in particular. The problem has got worse as the capabilities have got better. After all, dumb phones (feature phones) were pretty secure since you couldn't upgrade the software and the networks put a lot of effort into making sure that the networks themselves were secure. Smartphones come with power and flexibility. With great power comes great responsibility, as Spiderman was told. Responsibility for security is a major challenge. Mike announced a number of different products that interlock like pieces in a jigsaw puzzle to provide end-to-end security for IoT devices up to the cloud backend. The pieces of the puzzle were: ARM ® Cortex ® -M processors based on ARMv8-M architecture with ARM TrustZone ® technology IoT subsystem with ARM CoreLink™ system IP for fastest, lowest-risk path to silicon Secure SoC designs fortified by TrustZone CryptoCell technology Complete wireless solution with ARM Cordio ® radio IP for 802.15.4 and Bluetooth® 5 Cloud-based service for secure management of IoT devices via ARM mbed™ Cloud Optimized implementation on ARM Artisan ® IoT POP™ IP for the TSMC 40ULP process At the "thing" end of the IoT, these new M23 and M33 cores extend ARM's TrustZone security technology into M-level devices, solidly targeted at IoT. The CryptoCell-312 makes providing secure IoT devices a lot more painless. Perhaps the most notable part of the announcement was the cloud-based service for secure management of IoT devices. This is the first time, I believe, that ARM has entered the SaaS market. One of the big challenges with IoT devices is managing the provisioning and update of software, not just the security of the devices in normal use. This needs to take account of the complexities of the whole supply chain (chips, boards, boxes, retail suppliers, and so on) and not some mythical world where an trusted expert enables each device by hand. Taken together, the cores, POP, and interconnect make implementing IoT devices simpler. The security, radio, and mbed Cloud services make end-to-end security available without having to build everything from the ground up. If this results in more secure IoT devices, then it can't come too soon. See my piece Video Cameras: No Service for You for some recent news in the area. To wrap up, Mike told us about the second winner of the wearables for good. He pointed out that cholera doesn't kill you, it is the dehydration as a result of diarrhea. So one challenge is to prevent the spread among underage children by very low-tech solutions like hand washing (among doctors, too, an alarming number of doctors in the west do not adequately wash their hands). He pointed out that John Snow was one of the first "big data" analysts, plotting out where a cholera epidemic occurred in London in 1854, and stopping it by the low-tech solution of removing the handle of one particular public water pump. In a similar way, a second message from this winner is that innovation is not always about technology. The team started with the idea of having soap that the teacher would have and ensuring that the kids washed their hands. The assumption was that it would have some clever electronic device in it, but in the end that is not what they came up with. The soap was on a rope so it would go around the teacher's neck and not get stolen or lost. It contained die so that when it was rubbed on the kids' hands it would color them. The way to get rid of it was for the kids to wash their hands. Of course, this could then be confirmed by checking that kids hands were free of dye. Mike said that there was some soul-searching about awarding the prize to this product. But they went back to the rules. Low cost, check. Reliable, check. Innovative, check. Prize winner...check. Previous: Who Wrote the Book on Formal Verification?![]()
![]()
↧