ARC - platform, shlapform
You say platform, I say shlapform
- Let's call the whole thing off
By The Chilli analysts
Configurable processor intellectual property (ip) vendor ARC
International has just announced it's next-generation system-on-chip
(soc) development platform, targeting design cycles of six months
or less. The Chilli examines this announcement and explores
a number of areas that ARC must work on to add real value.
A "SoC development platform" is just a collection of existing
point ip, unless?
The ARC platform contains a risc/dsp processor, peripheral
ip, operating system, software development tools, protocol stacks
and device drivers. So far that's on a par with what most microprocessor
vendors (IP or device) can offer today via their third party
network. It is not a complete solution for the majority of embedded
applications, each one of which has specific requirements, reflected
in the hardware/software make-up of any potential solution.
An ip platform, as defined by The Chilli, is virtually a complete
chip design, including application, operating system and development
tools software, which is optimised, verified and certified to
relevant standards, and can be manufactured at any foundry by
the licensee.
Where's the value to the customer?
A platform adds value to a customer's activities by solving
their headaches (resource, skill set, cost, time and considerably
reduced risk of project failure). A platform vendor is expected
to know the end market, its dynamics, and provide customers
with a complete, compelling solution, optimised to their needs
- as if the supplier was a virtual product division of the customer.
The ARC platform might be a good start but needs to add considerable
value to its collection of point ip commodity blocks made up
of a processor, usb and Ethernet. The customer is required to
add application-specific hardware and software just to get a
baseline level of functionality and only then can the customer
add those features that differentiate the product from their
competitors. ARC, amongst other ip vendors, is effectively asking
customers to reinvent the wheel before focusing on their core
value-add. Giving a customer a few basic building blocks and
leaving them to work it out for themselves is a distraction
for them, and will make it more likely that the project will
fail or alternatively they will use another solution.
Nine months for a baby - six months for a SoC
Developing a SoC is a time-consuming business, involving market
research, specification, and development of hardware/software
modules, testing them individually and then testing them once
integrated. Shortening product cycles requiring parallel development
streams, outsourcing, complicate this further. Feature creep
adds yet more complexity.
To deliver a complete solution the customer must assemble the
various hardware and software components. They may have to design
or modify their own ip or procure it from elsewhere, often involving
several vendors. When the customer integrates this disparate
ip, he discovers what The Chilli refers to as "mutually
incompatible ip", that is, blocks from several different parties
that were not designed to work together.
System-on-chip products are growing in complexity, not just
in terms of gate density, but also in terms of bus architecture,
memory, clocking, mixed-signal functionality. Software is another
challenge to develop and integrate. Developing a solid system
architecture is tough; designing and integrating the various
hardware and software modules is harder. But this pales into
insignificance when you have to verify the entire design. One
tiny mistake can manifest itself as several different faults
across the product. The customer must then pinpoint the fault,
address it with the responsible party, and integrate a fix which
may cause problems elsewhere. Such an iterative process, involving
several different parties and much complexity is highly unlikely
to be completed within six or even nine months. Providing less
than half a solution is not going to reduce the verification
time significantly, especially if the processor configuration
has been altered.
Contrast this state of affairs with a customer procuring an
application-specific standard product (assp) with the necessary
hardware and software, including application-specific middleware.
This is the case for those vendors who have developed assp solutions
for markets including digital TV, mobile communications, etc.
Their success is visible in terms of the market share held.
Summary
To be frank, it appears that this platform announcement is
very similar to the previous platform announced in February
2001, apart from the change of processor from -A4 to -A5. If
this is the only improvement in eighteen months within ARC,
then it is hard to see how they can reduce the design cycle
for licensees unless they are re-focusing their effort towards
field-programmable gate array (fpga) design and away from SoC.
Tools and ip for fpga command very little value nowadays, as
the fpga vendor is more likely to give this away to encourage
a sale of high volume. However it is unwise to be unduly harsh.
If the ARC platform really can reduce the design cycle of a
typical SoC to below six months, then customers will flock to
evaluate and license it. Allowing for a six month development
cycle, the customer should be in production in around nine months
time, bringing ARC closer to realising a royalty stream [Editor's
note: royalties at ARC still only stand at £0.2 million for
Q3]. This would be good news for everyone. However, as a devil's
advocate, what is better for a customer - providing point ip
that requires them to reinvent the wheel before they can get
started, or a platform that allows them to start differentiating
themselves immediately? The Chilli contends that the
latter option is more likely to result in higher royalties,
sooner.
Future issues of The Chilli will examine the strategic
options that companies like ARC can adopt to get out of the
current dilemma faced by most ip vendors.
Comments on this story? Send an email to Woz Ahmed:
woz@thechilli.com
|
