As I did at the Design Automation Conference in Austin earlier this year, I will put a blog out with the highlights of the day (at least the ones I saw, I can't attend every track obviously). It's not as good as being here, and it also serves as a trailer for some later blog posts when I decide a subject is interesting enough to deserve a day to itself. Welcome The first day of CDNLive here in India, Tuesday, started with a welcome from Jaswinder Ahuja. I don't think I've met him since I left Cadence last time back in 2002. He gave the welcome. He said that there were actually too many people to fit in the largest ballroom and so there was a video feed to another area. It seems around 900 people actually registered and then turned up. Jaswinder said that there are four keynotes, two each day, along with 85 user papers. Keynote: Michal Siwinski The first keynote, Expanding Your Opportunities with System Design Eablement was given by Michal (actually pronounced MeeHao, Michael in Polish). I will cover this presentation another day, probably next week. The big trend that he pointed out was that market leaders optimize across multiple layers of technology to create differentiation. This is changing how designs need to be optimized, especially in the four big semiconductor opportunities: Mobile Datacenter/cloud Automotive (changing) moving into advanced nodes Industrial (analog, connectivity) great growth area in the future Increasingly, each of these markets has a lot of base commonality but also increasingly their own special requirements. This is perhaps most obvious in automotive with its unique safety and reliability requirements. Keynote: Dr Sankaran, Analog Devices Inc The second keynote was by Karthik Sanakaran on IoT Technogy and Application and Technology Drivers for Semiconductor Industry . Dr Karthik is the General Manager of IoT for ADI. He pointed out that it is a 51-year-old company and he describes what they do as "real-world processing", taking signals from the real world, doing analog system processing, digital signal processing, and conversion to the digital world. IoT falls squarely into this, since it connects the physical world to the cloud infrastructure. He thinks that there have been three big waves of semiconductor. The first wave was mainframes and similar computing. The second wave was the PC, the laptop, and the smartphone. The third wave is ubiquitous sensing. He also has a sort of Maslow's hierarchy of data, where data goes from raw data, to information, to understanding to wisdom. IoT does a lot of this since it senses, measures, processes, and then transmits. The simple way to do this is to upload the raw data. But that is too power hungry and bandwidth intensitive. The node (the "thing") needs to turn the data into information, which enables it to move from reactive to predictive with real-time decision making. One big area of opportunity is "condition-based maintenance". Instead of scheduled shutdowns for maintenance, IoT technology can be used to sense when problems are building up (or when they are not), which reduces the operating costs of the plant and improves the business process. In fact, IoT in business is largely about how much businesses can transform. Less than 1% of the data gathered is currently used, primarily for alarms. But by adding deep learning, the data becomes much more valuable. He had another example of a potentially transformative IoT product in SCIO, a Kickstarter-funded Israeli company that has a hand-held sensor that connects to the phone, uploads to the cloud and does a spectral analysis, compares it to a database, and then can tell you the material. Currently Kickstarter says that due to an intellectual property dispute the product is not available, so it might not have been the best example. But there are obvious applications from food, to health-monitoring, detecting fake pharmaceuticals, and more. The implications for IC and system design are: Low power: SInce it costs costs $100 to change a battery in US, designers want 20 year battery life for industrial sensor nodes Energy harvesting System-level optimization and insight: The biggest source of differentiation and impact, which is one of the major points Michal had made Increasing hardware/software interactions Security: Key to growth of IoT Newer architectures to support machine learning, both in cloud and at edge Since this was verification day, he talked a little about the trend to increase productivity and re-use everything across the development teams. He talked about the portably stimulus working group (PSWG) which is an area being driven by Cadence (along with Mentor). Indeed the picture that he used, which officially is from Accellera, looked suspiciously like Cadence's own Perspec slide that Michal had used earllier. Michal, Update on Verification Then Michal was back with an update on verification. I'll cover that when I cover his keynote, since they cover some of the same ground, just in different levels of depth. Invited Presentation, Open Silicon Next there was an invited presentation by Mitesh Thakker of OSI on Conquering System-Level Issues with LPDDR3 IP Flexibility . He has been involved in development of an SoC camera chip to be used a cross a range of products. It is a package-on-package device with memory and logic bonded together in a single package. One problem with that is that it is basically impossible to probe everything you want. They have a three-phase development process: Engineering sample or ES. There are a limited number of samples at various process corners on a debug board Preproduction or PP, a limited number of samples on a production board Mass production or MP, after which 1000s of parts are available. The part had a number of problems with initialization failure, read and write leveling failure, and so on. If you know anything about DDR memory interfaces, you probably know that the margins are now so small that unbalanced skew can occur. Consequently, the interface has to be trained to take account of different board traces, process corners, package pins, and so on. This training takes place on power up and usually on a timing-based intercal (since it is voltage and temperature sensitive). I want go into all the detailed work that Mitesh described to come up with theories as to what was wrong. Almost all the theories turned out to be wrong. It wasn't badly done timing analysis, problems with the DLL, pad calibration issues, nor issues with the leveling finite-state-machine. It turned out to be system-level jitter. The problem could be fixed by reconfiguring the LPDDR3 IP to allow more delay elements, the original choice of 1 or 2 was not enough and 2 and 3 were required. Another issue was failure of reads occasionally at subzero temperature. It turned out to be a crosstalk problem but the device was not large enough to add more decaps. But by making changes to some more parameters at the IP block, the problem could be solved. The bottom line message was that these were both system-level issues (jitter, crosstalk) that could be solved, not by fixing the root cause, but by making use of the IP configurability. Since this was a presentation at CDNLive, I assume that this is the Cadence LPDDR3 interface that is being discussed. Interview with Michal Next I was back with Michal. We had to do a video interview for an Indian automotive magazine. It was the first time I'd done an on-camera interview like that (the second will be tomorrow). The first one had some problems, mostly that we got to chatting and it went on for six minutes. The next time, we left out a couple of questions and raced through in under three minutes. Then we got the Goldilocks speed...except the hotel knocked on the door to say the fire alarm was being triggered, probably by the lights. But we had two cameras, so we picked up in the middle and it was a wrap. I'll post a link to the video once it has been edited and is online. One of the questions I had to ask (the magazine provided the questions) was about the "made in India" government initiative. I realized that with two large development groups, quite a lot of Cadence tools are "made in India." Previous: CDNLive Bengaluru, a Long Journey![]()
![]()
↧