IP Subsystems Are Nothing New

Arteris, 2011年04月28日

Rich was kind enough to send me the table of contents and an extract of the executive summary of this report, and
I will say that I agree with the conclusion: Instead of integrating discrete IP blocks, “SoC designers are now
looking to move up a layer of abstraction to design with system-level functionality to reduce the effort and
cost associated with complex SoC designs today.”

And Rich also states that, “the concept of an IP subsystem is not a new one.”

However, I am confused about why people think this is new, and whether our definitions of an IP subsystem go far
enough.

Are IP Subsystems New?

I’m totally biased based on my experiences working at Texas Instruments and ARC:
IP subsystems are not new.

  • 2002, TI’s OMAP 2 IVA 2: Back in 2002 at TI, we had the OMAP 2 processor platform with its IVA 2 image and
    video accelerator subsystem. This subsystem combined hardware with a software stack consisting firmware,
    operating system hardware abstraction layers/drivers/board support packages (for multiple OSes), and even
    middleware running on the OS.

But wait. TI is a silicon vendor, not an IP vendor. When has there been an instance of an IP vendor offering an
IP subsystem?

  • 2004, ARC’s ARCsound: As far as I can tell, ARC (now part of Synopsys) was the first to offer an IP
    subsystem with its ARCsound product in 2004. This audio IP subsystem was, “a preconfigured, licensable audio
    platform for use in high-volume SOC designs. The ARCsound subsystem includes state-of-the-art audio codec
    software, a pre-configured ARC processor with custom audio extensions and a full suite of hardware and
    software development tools.” (From ARC press release, “ARC International Introduces ARCsound: Licensable
    Digital Audio Subsystem, Sept. 27, 2004.”)

Therefore, semiconductor intellectual property vendors have been creating commercially available IP subsystems
for at least 7 years.

What should be in an IP subsystem?

Rich lists three features that are part of an IP subsystem, and I agree with all of them. Paraphrasing Rich, IP
subsystems:

  1. More closely resemble system level features rather than discrete IP functions.
  2. Can be sized to match the application requirements exactly.
  3. Have their own interconnect.

As an interconnect IP provider, I especially like No. 3.

Dude, where’s my software?

OK, I apologize for my homage to the worst movie ever.

Seriously, as I look at this list I see one missing item and a second potential missing one.

First, software is missing from the definition. Look back at my descriptions of TI’s OMAP 2 IVA 2 and ARC’s
ARCsound: They both included a software stack leading all the way up to the application level. That is because
as development times for consumer electronics have shrunk, the CE device vendors have put the screws on
semiconductor vendors, forcing them to supply enabling software for their silicon in addition to the chips
themselves.

Things like this roll downhill through the value chain, so an IP provider hoping to sell an IP subsystem had
better provide a complete software stack to the semiconductor maker.

Regardless what technology is, I like analog, too.

This heading is a quote by Foreigner lead singer Lou Gramm, who understood that we all interact with the world in
the analog realm, not the digital. Each IP subsystem will have to interact with the world through analog IP and
devices. Should these be part of the IP subsystem?

I don’t know enough to answer this question. But I do know that there is a trend toward offering IP subsystems,
and that software will be a key component of that subsystem.

Subscribe to Arteris News