As usual, Arm TechCon opened with a keynote by Mike Muller, Arm's CTO. His son is in mechanical design in some way and Mike told us he is applying for an internship. In a sign of the times, he is applying for an internship in Shenzhen. If you don't recognize that name, it is a city in China just over the river from Hong Kong, where a huge amount of the world's electronics is manufactured. If you have an iPhone, it was made in Shenzhen by Foxconn (or Hong Hai, as it is known in China). Emotions His son was putting together his design portfolio and told Mike about emotions being involved in design, and about Robert Plutchik's wheel of emotions. His wife told him he was very good at showing his emotions, not so good at talking about them. I think that just means Mike is male, but he decided to talk about emotions during the keynote anyway. He started with disgust, and showed a picture of a premature lamb in a plastic bag with tubes and electronics attached. It wasn't quite clear what the lamb was doing in a bag, but it was something to do with Crispr-cas9, which I'm sure you've heard about. It doesn't really have anything to do with electronics, so I've not talked about it (hey, there are some subjects that Breakfast Bytes doesn't cover!) but I'm sure you've heard about it since it's mainstream enough to have been on the cover of the Economist a couple of years ago. Disgust is pretty...well...disgusting. So Mike moved onto fear and surprise. Put them together and you end up with awe. Mike had a video of SpaceX's successful landing of their Falcon 9 rocket on a barge, in April this year. For those of us who were old enough to have watched the NASA Saturn V program, this is pretty amazing to see. Mike pointed out that Dipesh (giving the second keynote) would call the rocket an IoT device, and it is clearly an autonomous vehicle, and connected. No matter what the conference, it is an all-purpose video. (Please visit the site to view this video) Machine Learning The next emotion was anticipation. Machine learning is bringing an extra dimension to what can be done. Historically, machine learning has bene in two pieces, training and inference. But it is becoming more distributed, in particular with a new level between the cloud and the edge, the gateway. This can vary from a something in your home, to a cellular basestation, or some sort of industrial hub. This is all blurring the lines, with some training, especially incremental, being done at the edge or in the gateway. There isn't enough bandwidth or power to get everything up to the cloud for processing, plus there are issues around security and privacy that make keeping your data on your device desirable. Next, Mike had some sobering statistics. One in five of us will have a heart attack, it is the most common reason for over-65s to visit the emergency room. It turns out that swollen ankles are a symptom of danger. He showed an odd-looking IoT device that would detect that and generate an alert. You put it near the floor in your bathroom and forget about it. When you are walking around in your bathroom, it is quietly checking your ankles. Inside, it consists of a few Raspberry Pis, a cellular dongle, and a hard drive. Even like that it can be deployed in thousands of homes economically. To deploy in millions, it could be integrated more and cost-reduced. But one reason Mike is confident that IoT will be transformative, and that the huge numbers of IoT devices forecast will, indeed, be deployed, is that there will be huge numbers of "cut and paste" products like these. Not every IoT device requires its own special chip. CPU Verification Mike moved on to CPU verification. One thing he mentioned is that Cadence and Arm had announced Xcelium simulation running on Arm servers that very morning (see my post Xcelium Simulation on Arm Servers for details). He had some statistics on verification of the Ares CPU (I think that's the Cortex-A75). It took 97 years of people, and 1500 years of CPU time. Just looking at the load/store unit, which is about 15% of the processor, generated 156GB nightly, for analysis the next day. The problem is that the first few tests get a lot of coverage. Most of the pipeline is exercised after running a few instructions. Then not much. Come 3am, suddenly a few tests hit something new. In the morning, people look to see what was not covered and add another test the following night. Arm has looked at whether you can train a machine-learning algorithm to distinguish good tests and bad tests. Generating tests is cheap but running them (repeatedly) is expensive. So if you generate thousands of tests and then use machine learning to pick the good ones, you can perhaps do a much better job. They used 15% for training and then ran on the other 85%. Some parts of the design had no coverage in the training set, but improving poorly covered parts improved them. As a result, they could make verification go twice as fast, which they could then take as reduced costs, or improved quality, or shorter time to market. Security, Sadness, and Joy Next emotion was trust. Mike talked a lot about Arm's new security architecture PSA (for Platform Security Architecture). In the press Q&A later, someone asked them why they named it PSA, something associated with bad experiences at the doctor. I've probably just lost half the audience here, since PSA stands for prostate-specific antigen, something women obviously don't require (they have their own annual indignities at the doctor). I wrote about the non-medical PSA in detail in Putting the Bad Guys in an Arm Lock . Next emotion, sadness. Mike just passed over, since it was too sad, although he did point out that David Bowie's name is pronounced Bough-ee, not Bow-ee, something Americans mostly get wrong. Finally joy. Mike had a Nokia mobile phone from 20 years ago, 1997. Not a smartphone, of course. The mobile industry in general, and Nokia in particular, had done a lot to make Arm the success it is today, which obviously brought Mike plenty of joy. Speculating about the future he thinks is going to bring technology together with some of the "wetter" technologies, like that premature lamb in a bag that he opened with. He thinks that in 2037 his son will look back and see that we have turned the impossible into the possible. Arm TechCon 2018 Arm TechCon is on the move. Next year it will be in the San Jose Convention Center, October 16-18, 2018. Presently, those are also the dates for the Linley Processor Conference, but he is here at Arm TechCon and I told him this and he said that is obviously a problem. So expect the Linley Processor Conference dates to change if you are ahead of the curve and already have those dates in your calendar. Sign up for Sunday Brunch, the weekly Breakfast Bytes email.![]()
↧