New Deep Learning Processors, Embedded FPGA Technologies, SoC Design Solutions

#55DAC: Must-see technologies in the DAC 2018 IP track.


Some of the most valuable events at DAC are the IP Track sessions, which give small and midsize companies a chance to share innovations that might not get much attention elsewhere.

The use of IP in SoCs has exploded in recent years. In a panel at DAC 2017, an industry expert noted that the IP market clearly was growing even faster than EDA itself, due to the fact that more and more chip makers are externalizing IP development functions. This is similar to what happened with EDA tools in the 1990s. The externalization trend is ongoing, and there is still more IP to be go. With continued advances in semiconductor processing, new systems-on-chip (SoCs) are always increasing in complexity. Research firms estimate that some SoCs include up to 200 blocks of IP, licensed from multiple companies, ranging from an I/O cell all the way up to an Arm CPU. Identifying and acquiring or designing all of the IP blocks is just the beginning. The big challenge is when you have to bring all of that IP into one design database and manage the versioning and design flow requirements of each unique block.

So if you’re working to optimize your designs with IP, come check out the DAC IP track this year. Most of our sessions consist of invited presentations and panels. We have nine 90-minute sessions total, from Monday through Wednesday of DAC week. Five of the sessions are for invited presentations—three 30-minute presentations each—and two for panels. The final two sessions will showcase submitted and reviewed presentations from industry and academia. This year, the submissions seem to reflect the increased interest in IP-based design, as we received 25 percent more than we did last year.

One area that is hot across DAC this year is machine learning, perhaps no surprise given how much artificial intelligence and machine learning technology is proliferating in systems and in the media. We have an IP for Machine Learning session that will be fascinating. Cambricon, an up-and-coming Chinese company, has created the first AI deep learning IP in the world. On Monday morning, they will present the details of their deep learning processor. In the same session, you’ll hear from Cadence, which has developed a neural processor based on its Tensilica architecture, and Dr. Vivienne Sze from MIT, who will discuss the types of hardware that are required for deep learning.

The IP track also includes two panels—one of which focuses on the perennially “hot” topic of low power. One panel includes representatives from TSMC, Microsoft and Minima, and will focus on the latest problems in achieving low power for new applications such as virtual reality wearables, IoT devices and even automotive systems.

The IP Track sessions are also a good opportunity for engineers to learn about some types of IP that most have not gotten a lot of exposure to yet. We have an excellent session on embedded FPGAs that includes presentations from some of the leaders in this new market: Achronix, Flex Logix and Menta. They will talk about three very different ways of implementing FPGA as an IP which can be instantiated into an SoC, as opposed to the other way around (like Xilinx’s Zynq or Intel/Altera’s SoC FPGA line). Embedded FPGAs used to be a tough sell because FPGAs traditionally required a great deal of area, but eFPGA vendors are getting traction now with these new offerings which are purpose-built products for embedding in SoCs, rather than being derived from existing full chip FPGAs. I’m interested to find out what has made it possible for these vendors to go from “it’ll never work” to “yes, we have customers.” In fact, Flex Logix will have Sandia Labs, a customer, co-present and describe the end application which takes advantage of the reprogrammability of the eFPGA.

Another session I’m personally interested in deals with IP to support CMOS image sensors (CIS). The CIS market is exploding, and image sensors are being integrated in SoCs for multiple applications, including automotive. These new image sensor IPs require new architectures and also innovative mixed-signal design. Chronocam, Austria Micro Systems and Forza Silicon will each present their solutions in this space on Tuesday morning.

At the conference this year, we have an excellent mix of EDA, IP, SoC, and foundry process engineers who will talk about how our industry is changing. This is going to be a great year for DAC. The event is back in San Francisco again this year, so it’s an easy hop from Silicon Valley, just grab the bus and come on up! I look forward to seeing you in June in San Francisco (June 24-28)! Register today for the Designer and IP track badge.

Leave a Reply

(Note: This name will be displayed publicly)