The other striking thing is that semiconductor IP gives Cadence the
opportunity to be paid a percentage based on the success of its customer
through a per-unit royalty. The semi-IP business grew 11.2 percent in
2012 according to Gartner Inc. while the overall semiconductor market
fell back by 2.7 percent according to the World Semiconductor Trade
"Four years ago we formed a partnership
with ARM. We listen to the customer and they are ready to outsource
their IP," said Tan. Indeed there is an argument that it is only by
steeping up to a higher level of abstraction and assembling sub-system
blocks or optimizing whole system-chip platforms, that engineering teams
can get their SoC's made on time.
The mention of partnership
indicates that Cadence would be unlikely to tread on ARM's toes by
getting into mainstream CPU or GPU cores or even fabric IP. "We are
selective in terms of what IP we pick. It has to be high-quality. Denali
IP is very strong and the operating margin is in the mid-20s [of
percent] so it is not subsidized business. Similarly, I have known the
Tensilica team for many years. We made sure it is compatible with ARM
because ARM is a very important partnership," said Tan.
reason that Cadence only wants to supply high quality IP is because that
is what it takes to command a per-unit royalty, said Tan.
aspect of the IP model also appeals to Tan; that it can be a repeat
business. For communication standards such as Ethernet, USB and MIPI and
others, the standard changes continuously and the optimization for each
process node may also change. When customers sign up with Cadence they
are often signing up a road-map deal whereby Cadence engages to bring
the most up-to-date functional implementations at each process node. "We
have a road-map and customers look at that road-map," said Tan. Some
customers could not engage with a privately-held core IP provider in
this way for fear the company would disappear or get bought by a rival.
Cadence represents a safe home for such IP road-maps, said Tan.
addition the library of IP cores can be characterized and optimized for
Cadence design flows and vice versa so that these two elements of
Cadence's business mutually support each other. This is an advantage
that chip vendors who provide IP, such as Xilinx and Altera, and
foundries, such as TSMC, do not have.
Click on image to enlarge.
Where Cadence's IP factory creates value. Source: Cadence.
Let me just comment from other perspective, cause "the bigger, the better" is not always true in IP market. Many times stuff from the 800-pound Gorilla IP vendors is not as much innovative and reliable as from smaller, but highly specialized IP providers. Just let's take as an example, which company has the most advanced IP solutions eg in 8051? And what about other IPs?
I take your point.
Most of the innovation in IP has been done at small, privately-held companies. After they get bought by Synopsys or Cadence does the innovation continue?
But Lip-Bu Tan's point is also valid and has been true for EDA tools vendors for a long time as well as IP cores.
Larger companies do not like to depend on small privately-held companies in case they go bust or get bought by the competition.
By the way who do you think produced the most advanced 8051 cores? Another Polish IP provider perhaps?
Is 8051 core a bit old-fashioned?
Thanks Peter for taking my point. As I mentioned, I don't say that being bigger is something wrong, but in many cases "big structure" can kill or at least weaken the spirit of innovations.
And re 8051, you're right, there's another Polish IP provider, who introduced the DQ80251 IP Core. Quality confirmation for this solution came not only from customers but you'd probably heard about EDN's Hot 100 products of 2012. Of the many thousands of products announced during the past year, EDN has chosen 100 that especially caught the attention of editors and readers. They range from ICs and components, to software, test instruments, development tools and sensors, and more. http://www.edn.com/electronics-products/other/4401454/EDN-Hot-100-products-of-2012--Microcontrollers--processors-and-programmable-logic . So would "a bit old-fashioned core" awarded by editors and readers of EDN?
I believe so. We cannot stop civilization process and transition to 32 bit MCUs is unstoppable, but still there's a place for other, innovative solutions - no matter if they are 8051 or 68k. Mentioned earlier DQ80251 can compete with eg M0...
There's also something more: http://www.soccentral.com/results.asp?EntryID=40462
It shows clearly, that there's a lot of place both for 8/16/32 and 32-bit MCUs - as long as they're innovative.
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.