Travelling On The Silicon Road

Coming close on the heels of ARM’s CEO, Warren East’s remark that ARM based servers will be out in the market in the next 12 months,  comes news from Marvell – the fabless company aims to supply silicon for ARM-based servers with 40-nm multi core processors that it will ship this year. As per EE Times, Marvell and ARM are working with “multiple Tier 1 companies” to build larger trial deployments to validate ARM as a server platform. Separately, Marvell has opened up its dual-core architecture, drivers and board support package so partners can create server software needed to support the new market initiative”

Opinions about ARM’s foray into server market making a dent in Intel’s server biz vary. ARM’s marketing chief, Ian Drew recently pointed out that “We’re an [intellectual property] company — we learned in last two years that we don’t know everything about everything. On the servers, it will take longer than we imagine because it’s a much more complex problem [than smartbooks or phones]”. ARM is running one of its websites (for the ARM Linux Internet Platform) on a cluster of ARM based chips, part of a handful of experiments to test out the viability of using its chip architecture in severs. They are Marvel based.
 

The server market, long dominated by Intel, has lately seen increasing power concerns. Google’s recent acquisition, Agnilux, was rumored to be making servers with ARM architecture. TI is researching the use of DSPs in servers. Intel’s best hope can lie in working along with companies using its low power Atom chips to build power efficient servers.
 

“East wasn’t willing to name the chip foundries that ARM is working with, however he said that information should tip up in the next 12 months.” Now if one of these foundries happens to be GlobalFoundries, it’d be interesting to see how the server market pans out – Intel’s long term rival, AMD and Global Foundries share the same parent company.

Intel will port unspecified Atom processor cores to TSMC’s technology platform, including processes, IP, libraries and design flows under the terms of an agreement between the two companies announced yesterday.
 

The deal will expand the TAM for Atom. It allows TSMC & Intel to go after newer market segments together - namely the embedded, CE, netbooks and handheld market.

Some interesting highlights on this deal:

• It is indeed a rare event for Intel to allow its processor to be manufactured by another company.
• Intel has no plans to license or transfer its high-k process to TSMC. It is doubtful the TSMC Atom devices will include high-k and metal gates. Hence it’s unlikely that the core from TSMC will have the same performance level as that from Intel.
• With PC shipments failing, Intel needs to aggressively penetrate into other markets.
• Intel goes head-on with leading embedded core provider, ARM. While ARM is trying to move to larger devices (from handsets etc.) for its cores, Intel is moving in the opposite direction (PC to smaller devices)
• We may see two strong foundry-IP camps emerging: ARM-IBM alliance and Intel-TSMC for the much sought after mid size converging devices in the embedded space.
• Point to be noted is that while there are 2 different market segments for the processor – higher end processors for PCs and the lower (& cheaper ones) for MIDs, Intel will need to do a real balancing act here so as not to have its profits from the upper segment being eaten away substantially by the lesser profits of the lower end processors.
  

Denali Software is on the trend of IP providers morphing to provide a complete solution/platform instead of standalone IP blocks. Starting with an offering of memory modules, then adding on semiconductor-related memory IPs to its portfolio, it has now introduced FlashPoint -  a full PCI Express interface platform. The platform includes a complete hardware design ready for fabrication plus a complete software stack and all necessary drivers.

From offering all the separate IP components and related software & firmware needed, a platform for PCI Express seems a natural progression - all the more logical and practical to survive the wave of consolidation.

There is an interesting article on ARM vs. Intel competing in the lucrative evolving product category placed in the gap between the smart phone and the laptop computer.

Intel looks set to invade the low power handhelds – a traditional domain of ARM. And its strategy to this effect involves offering X86 processors with leading performance while reducing the power consumption and footprint. ARM meanwhile continues to come up with various processor architectures and cores optimized for its customers’ needs and semiconductor manufacturing capabilities of its licensees.

The question is: Will future mobile handhelds run on an X86 or ARM instruction set architecture?

I summarize here a few points for both parties:

Advantage ARM:

• The ARM processor was built for mobility from the ground up
• ARM’s strategy looks to be targeted more on “needs of its customers” vs. Intel’s “what it can do with technology”
• ARM works along with partners to develop an ecosystem where various entities contribute by building the system with ARM’s core while retaining their differentiating features; vs. Intel’s strategy of setting out a standard, defining the platform architecture and letting others build upon that. The latter has worked for PC and laptop market but may not get the same success in the volatile, different features for different people consumer market.
• ARM’s TrustZone technology which gives an ARM CPU a secure state

Advantage Intel:

• It is big & its pockets are deep
• It has the silicon manufacturing edge
• It can help OEMs with their branding & marketing
• Apple may be leading the way in this market and Intel can buy its way back into telephone silicon

The coming months will tell how flat the world has become…… 

Fuelling the ambition to become a leading provider of chip design services and IPs i.e. in addition to the top post of pure play foundry, TSMC has acquired Ottawa’s memory IP start-up, Emerging Memories Technologies Inc. (EMT).  

EMT, a start-up with a relatively small group of memory technologists (and headed by 39 year old Sreedhar Natarajan) specializes in the design and licensing of leading-edge embedded memory technology in both bulk CMOS and SOI and was launched in 2004 when the memory design biz of Atmos Semiconductor was taken by Mosys. Many employees at Mosys moved to EMT – and now will move to TSMC.

As I noted in an earlier post on a similar topic in May this year - call it as seismic changes or consolidation, the chip manufacturing world is going through some upheaval.

Call it as seismic changes or consolidation, the chip manufacturing world is going through some upheaval. While on one end, quite a few IDMs are transitioning to a fab lite strategy (albeit in different flavours) - especially with the high costs & risks involved in sub 65nm, on the other end heavyweights like TSMC are spreading their reach into the IP arena too.

As I noted in an earlier post, TSMC is leveraging on its resources and market reach.It may start off as “strengthening the design collaboration for critical sub circuits” (who doesn’t want FTSS??) but the intensity of the move has been enough to spread ripples in the till now independent IP biz world. By doing so, TSMC may well be doing its share in mitigating some of the risks involved in DSM design and thus catalyzing more of these design starts; and subsequently fill up its high-end fabs.

TSMC is broadening it’s portfolio of internally developed IPs and 3rd party IPs. It had started a program called IP-9000, later renamed to Active Accuracy Assurance Program, to qualify various IPs in its foundries. The objective was to expedite the design time with silicon proven 3rd party IPs.

With shorter design cycle time and with IPs becoming mandatory blocks in a design, the need for silicon proven IPs is not just desirable but also essential. Having a broad and quality IP portfolio is a big asset. If TSMC is getting into the ASIC like biz model, then indeed it is worrisome for the 3rd party IP vendors; especially the smaller ones who aspire to gain market share on the basis of their expertise in niche areas. The field gets all the more “unlevel”. But then it is a competitive world and TSMC would be leveraging on its resources and market reach.

A point to be noted is that, does this mean the resurrection of ASICs - often ranted about as dead ??

As pointed out in the article, most designers treat IP as a product. However, this product rarely comes with a guarantee; which is not that surprising. It’ll be almost suicidal given the argument that IPs are not plug in objects. Not only the IP’s functionality but also its interface and integration with the other components in the system determine whether the chip will work or not. And a standalone guarantee for an IP does not hold much credence.

A close working relationship between IP supplier and user has always been deemed vital for the successful IP usage and integration; hence to formalize it and bundle things under the “service” umbrella will not be that major a leap of faith.

To sell wafers, one needs tapeouts. Successful tapeouts require libraries and IPs validated on the target technology. And as technology advances, customers are more and more wary of getting their designs taped out with libraries and IPs not fully validated on silicon. So where does this lead a foundry with a ready advanced process but waiting for library & IP vendors to provide their wares on this new techno so that it can get customers’ designs in ?

Well, it provides libraries and IPs - either on it’s own or with partnerships. TSMC’s Europe Technical Director, Douglas Pattullo said in the IP/SoC conference in Grenoble on 7th Dec – TSMC is a provider not just of libraries but of complex IPs as well. He mentions that they are doing it to support their wafer manufacturing biz and not to get a new revenue stream.

It was once the same with EDA vendors. Quite a few of them started providing an IP portfolio – yes, to support their EDA biz. After all, customers are more comfortable with 3rd party IPs proven to be working in a specific design flow. But then as the process world started getting interleaved with the design world & the design space became abuzz with terms like DFM, DFY etc., the impact of foundry information on the EDA and IP space gained further importance.

So, are we headed towards a landscape dominated by a few major players (with deep pockets & partnerships) sporting One-Stop-Shops & dotted by smaller players excelling in niche areas say point EDA tools, special IPs ?