There's been much talk about the tools and methodologies needed for next-generation electronic systems design, but not so much about the people behind them. The people side of system-level design became clearer at a DesignCon panel titled "Who is the Designer of the Future?" One conclusion: RTL designers are going to need some new skills.
This Cadence-sponsored panel was chaired by Ron Wilson of UBM Electronics (EE Times and EDN). Panelists included:
- Ashok Mehta, senior manager, TSMC
- Gary Smith, chief analyst, Gary Smith EDA
- Grant Martin, chief scientist, Tensilica
- Geoff Burns, VP of product management, Silicon Hive (provider of image and video processing IP; see our recent Digital Implementation blog for more information)
The panel was heavy on "ESL" (Electronic System Level) expertise, and the discussion focused on the shifting roles of system architects, software developers, and digital hardware designers. I think the "designer of the future" in the analog/custom world would be a good follow-up discussion.
New Skills for a New Methodology
Panelists first gave their opinions about the design methodology of the future. All agreed with the need to move above RTL. Mehta said that move to transaction-level modeling (TLM) will be crucial for software development, bandwidth and latency issues, dynamic power management, and functional verification. Martin and Smith talked about platform-based design, in which pre-developed silicon IP may take 80 to 90 percent of the silicon area.
So, what type of people will use this new methodology? Hardware people who know a little about software, software people with some awareness of hardware, systems architects, or a new type of designer altogether? Burns said that experts in the system's target applications will need to partition the design. After that, it's time to bring in the hardware designers to develop hardware IP blocks.
Smith said that a new "level" of designer is appearing below system architects. These designers have both hardware and software skills and are "not in the hardware or software bucket. I don't like to call it software anyway, I call it modeling. It's a new area of engineering."
Martin noted that today's designs offer a wide variety of choices, such as the configurable processor technology his company provides. Hardware/software partitioning requires "people who can cross the [hardware/software] boundaries and who understand something about the end applications and the algorithm, and can make choices like, do I want to extend a processor or design a hardware block?"
Mehta said that algorithmic model develop requires "software type people" who can design a very high-level platform. "But after that you need a hardware architect type," he said. Hardware expertise is also needed to migrate to RTL. "At the end of the day you can't tape out ESL. You still have to go to RTL."
End of the Road for Traditional RTL?
It thus seems clear that hardware designers will always be needed. But what about the conventional RTL designer? The message: get ready to learn some new skills. Mehta said RTL designers will have to learn to write synthesizable SystemC code, and learn how to partition designs at the C language level.
"If I were an RTL designer today, I would take a look at what's happening in verification," Martin said. "People who understand the verification side as well as the design side are likely to have a wider choice of career options over time."
Smith said the RTL design is currently being automated. "For all practical purposes, it's going to be push-button, and there won't be much need for RTL designers except in semiconductor companies." If you're not near the end of your career, and you're not at a semiconductor company, "you've got to shift up to ESL," he said. Burns, however, commented that "there are still problems where somebody who wants to stay in RTL can contribute."
I doubt that RTL design will completely disappear, but there seems little doubt that the designer of the future will need to understand end applications, cross traditional hardware/software boundaries, and focus on IP integration in addition to design creation. That's all very consistent with the EDA360 vision. The interesting question is how we train and educate these designers. It's going to take an entire industry, with help from public institutions, to solve that problem - and it's time to get started.