News & Analysis
Comment
KB3001
That, and also the business model of this new start-up. Would they adopt the ARM ...
jaybus
I'm confused. Why is everyone calling it a "faster, better, cheaper, lower ...
China startup rolls 'unified' Android processor
Peter Clarke
6/8/2011 5:47 PM EDT
LONDON – ICube Corp., a two-year old fabless startup out of Shenzhen, China, backed by a publicly listed holding company in Hong Kong, is going after the Android mobile computing market using its own architecture of multicore processor.
The company claims its Harmony Unified Processor architecture is the first ever developed to handle both logic and graphics processing in a single core and in each of multiple processing pipelines.
ICube Corp., now a subsidiary of the recently renamed ICube Technology Holdings Ltd., has received 65-nm silicon of the first instantiation of Harmony, the dual-core IC1, back from its foundry supplier. The company is aiming at a 2012 roll out of samples and volume production.
Any attempt to introduce a novel instruction set architecture to compete against the broadly-supported ARM, x86 from Intel and others, is likely to be treated with skepticism by many. Objections are likely to be on commercial as much as on technical grounds. However, ICube claims that starting with a blank sheet of paper and specifically targeting the needs of mobile devices, has given it power-efficiency and cost-efficiency advantages over the established architectures. It also claims that with the advent of Android the barriers to entry for a new architecture are much lower than they once were.
The executive credentials of the company suggest that it could at least have a chance of success. Chief technology officer Simon Moy has 20 years' experience having spent several years as a principal engineer with Nvidia Corp. where he worked on the hardware design of the vertex shaders and stream processors in several generations of GPUs. Prior to that Moy worked at Silicon Graphics, IBM and LSI Logic.
Chief scientist Fred Chow's expertise is on the compiler side where he has 30 years experience, having worked as principal engineer at MIPS, chief scientist at Silicon Graphics and director of compiler engineering at PathScale Inc.
"Developing our own intellectual property from the ground up has been a key differentiation of ICube from other technology companies in China," said Moy, in a statement. "This provides us with much greater areas for innovation, enables us to leverage the latest semiconductor trends and lowers the cost of our system-on-a-chip products relative to other chipmakers. We are also proud to have attracted and built a world-class development team composed of Silicon Valley veterans and top engineering talents in China. We are confident that our revolutionary high technology product will be well received by the market and create value for our shareholders."
More information on ICube and the Harmony Unified Processor architecture are due to appear in the next issue of EE Times Confidential. Further details of the China fabless chip company sector, including a database of company details can be found in the 'China Fabless Profile' report from EE Times. If you are interested in receiving the report, contact peter.clarke@ubm.com
Related links and articles:
www.icubecorp.com
www.icubetech.com.hk
News articles:
China link helps MIPS go mobile
Chinese IC developer uses ARM's Osprey for laptop designs
China's Godson MPU aims for 28-nm
Report: China group mulls stake in MIPS
|
|
|
|
|
Navigate to related information
Les_Slater
6/8/2011 10:20 PM EDT
This is quite impressive. China has the resources and much of the infrastructure to create a formidable center of technical innovation.
Sign in to Reply
Neo1
6/9/2011 4:52 AM EDT
A core with GPU+CPU tailored for Android? Too much of a gamble in a highly competitive market. What kind og gains in power saving or unit cost of these cores would make software houses adopt it. Very tough I would say unless they are planning to enable software development and spend big $$ on developers to at least stand a chance.
But interestingly, this core might find excellent uses in other niche domains like automotive and medical.
Sign in to Reply
bikebivvy
6/9/2011 5:55 PM EDT
Surely the software houses don't need to adopt it, they already do. It will run all android apps without alteration. They are writing a compiler because a big part of android is coded in C++. Sounds an exciting project, esp. for hardware engineers and compiler guys, to try to make it as efficient as possible. Android phones are all about cost - so it it has every chance of doing very well indeed I would say. Android apps are coded in Java, so Android programmers will not need to touch this chip, or the ARM for that matter.
Sign in to Reply
peter.clarke
6/9/2011 4:58 AM EDT
@Neo1
I just wanted to reiterate that this is NOT a GPU and CPU in a single core. It is a single ISA that handles both logic processing and graphics rendering.
As to the power-performance-cost benefits, these will be discussed a little further in a EE Times Confidential article. As to whether they are a enough to make customers adopt it -- the market will decide.
Sign in to Reply
CharlieCL
6/9/2011 8:57 PM EDT
That is interesting.
Sign in to Reply
dirk.bruere
6/9/2011 3:13 PM EDT
Looks like Linux has finally arrived
Sign in to Reply
DCSchmidt71
6/9/2011 3:27 PM EDT
What makes the Android market work is applications that run on ARM. Is this CPU going to be ARM-compatible? If not, it will be a captive application market, and consumers will shun it.
Sign in to Reply
Code Monkey
6/9/2011 4:02 PM EDT
They could emulate the ARM ISA and handle API calls natively.
Sign in to Reply
peter.clarke
6/10/2011 11:07 AM EDT
They could but I believe they are going to offer their own C/C++ and graphics compilers
Sign in to Reply
bikebivvy
6/9/2011 6:19 PM EDT
Simply not true. As Pizmek said, none of the apps run on arm, it is all java based. So it looks like this is a very nice idea indeed, as all games etc should run on it, but faster, cheaper and using less power. ARM will have to come up with something similar and soon.
Sign in to Reply
cdhmanning
6/9/2011 11:54 PM EDT
No. The Android applications run on a Virtual Machine called Dalvik. See: http://en.wikipedia.org/wiki/Dalvik_%28software%29
This virtual machine can run on any silicon. It does not need to be ARM.
Sign in to Reply
peter.clarke
6/10/2011 11:06 AM EDT
This CPU is "independent" so not an ARM ISA, it is claimed.
Sign in to Reply
przemek
6/9/2011 5:01 PM EDT
@DCschmidt71: Android is designed on a Java virtual machine, so android apps should port easily. In fact I always wondered why one of x86 vendors didn't make a droid (based on Atom or VIA/centaur)---the only reason I could come up with is that they aren't competitive on a combination of price and low power. I have actually heard from VIA that they aren't interested in low price points.
Sign in to Reply
cdhmanning
6/9/2011 11:56 PM EDT
Android has indeed been cranked up on x86 and MIPS.
The real reason nobody has sold an Atom based phone is that it would use far too much power. Unless you wear cargo pants you won't have the battery life to handle a whole day.
Sign in to Reply
przemek
6/9/2011 5:03 PM EDT
@DCschmidt71: Android is designed on a Java virtual machine, so android apps should port easily. In fact I always wondered why one of x86 vendors didn't make a droid (based on Atom or VIA/centaur)---the only reason I could come up with is that they aren't competitive on a combination of price and low power. I have actually heard from VIA that they aren't interested in low price points.
Sign in to Reply
przemek
6/9/2011 5:07 PM EDT
@DCschmidt71: Android is designed on a Java virtual machine, so android apps should port easily. In fact I always wondered why one of x86 vendors didn't make a droid (based on Atom or VIA/centaur)---the only reason I could come up with is that they aren't competitive on a combination of price and low power. I have actually heard from VIA that they aren't interested in low price points. A new Chinese entrant with low fixed costs could provide a challenge to everybody including ARM---of course provided that they can be similarly competitive on the rest of the system (peripherals, memory, screen).
Sign in to Reply
Frank Eory
6/9/2011 7:31 PM EDT
Sounds like they have a very impressive engineering team, but wow do they have an uphill battle in the market!
And 65 nm is fine for a test chip to have something to play with, but even being based in China, there is no way they will be able to compete on price at that node.
Sign in to Reply
Les_Slater
6/9/2011 7:40 PM EDT
That's why in my original post I didn't mention ability to manufacture.
What node is China at internally? Do they have access to foundries outside mainland that will process wafers for them?
There are political and economic as well as technical issues here.
Sign in to Reply
junko.yoshida
6/9/2011 8:48 PM EDT
These days, Chinese fabless companies do have access to foundries outside mainland, including TSMC. The situation has drastically changed in recent months.
Sign in to Reply
bikebivvy
6/9/2011 7:48 PM EDT
They could get the chips made outside of China and the rest of the phone in China. I cannot see that being a problem. It's a flat world now you know! What they will be doing is using a better chip that they have exclusive access to - to sell phones - and ones that will be fully compatible with existing Android apps.
Sign in to Reply
kagayl
6/9/2011 8:34 PM EDT
People seems to forget that Linus Torvalds was at Transmeta at one stage - a VLIW processor that can dynamically compilers x86 and 68k instructions. SO if the Chinese CPU can dynamically compile Java byte code, that would be it.
Sign in to Reply
KarlS
6/10/2011 9:21 AM EDT
It may not be necessary to even use the byte code -- parse the source then design hardware that evaluates the if clauses and does the appropriate assignments. I have done it for C statements. Also, using the byte code pretty much is done by a JIT compiler at run time so why can't they write a new JIT for their architecture and use existing compilers?
Sign in to Reply
abraxalito
6/9/2011 9:28 PM EDT
An example it seems of doing the design right but not doing the right design. Technical wonder but marketing flop.
Sign in to Reply
cdhmanning
6/10/2011 12:07 AM EDT
Explain why this will be a marketing flop.
Just remember that China does not need to sell these anywhere else to make a worthwhile effort.
China has a large enough domestic market to make something like this work, and work well. Once they've sorted that they can go for external markets.
Sign in to Reply
abraxalito
6/10/2011 1:20 AM EDT
Thanks for asking - very glad to put more meat on the bones.
Firstly marketing is not something that can be 'bolted on' at the end of a design cycle - as you see to be assuming. You wrote '..will be a marketing flop'. No marketing has input right from the start in terms of what the specification is and what customers needs are. So it has already flopped in marketing terms because in the story told by Moy customers don't figure very highly. Where are the customers clamouring for what he's offering? I see none - he's left it too late now to 'push' the offering into the market. Successful marketing results in anticipatory customers pulling products out.
That's just for starters - if you want more please ask.
Sign in to Reply
jelvidge
6/10/2011 3:15 AM EDT
"ICube claims ... power-efficiency and cost-efficiency advantages over the established architectures". This is easy to claim but requires data to substantiate. I will, as the saying goes, believe it when I see it. (And I don't just mean the PowerPoint!)
Sign in to Reply
abraxalito
6/10/2011 5:00 AM EDT
Not only that, but because existing architectures are indeed well established they need to show a compelling benefit to dislodge them. A few tens of percentage points will not suffice. Grove's Law would postulate they'd need around 10-fold improvement to disrupt the incumbents.
Sign in to Reply
Les_Slater
6/10/2011 5:19 AM EDT
I don't think marketing is such a serious issue here.
The primary issue is the competency of the team behind this project and the supporting infrastructure available to them in China.
I said earlier however that economic and political issues were part of the equation. In that respect it might be worthwhile reviewing Andy Grove's take on the subject: http://www.citizenstrade.org/ctc/wp-content/uploads/2011/05/20100701_howtomakeanamericanjob_bloombergopinion.pdf
Sign in to Reply
peter.clarke
6/10/2011 7:51 AM EDT
@jelvidge
As mentioned above, a more detailed article is slated for the next edition of EE Times Confidential.
Sign in to Reply
wave.forest
6/11/2011 7:36 AM EDT
Good luck to ICube.
I reckon its success chiefly depends on how they can sell the architecture to some big Chinese telecom OEM's, like Huawei and ZTE. If they can make it as a "State will", it definitely will succeed. Remember TD-CDMA and TD-LTE? All the major telecom OEM's are licking their ass.
Sign in to Reply
jaybus
6/11/2011 7:58 AM EDT
I'm confused. Why is everyone calling it a "faster, better, cheaper, lower power" processor at this stage? What evidence is there that any of the above is true?
Sign in to Reply
KB3001
6/11/2011 6:53 PM EDT
That, and also the business model of this new start-up. Would they adopt the ARM soft IP model? It does not look like it, in which case they have a mountain to climb.
Sign in to Reply