Friday, July 11, 2014

Value Proposition and the Custom Chip Market: Part 4 - the SoC Sector

"He went to Broadcom."

This phrase was probably the lead cause of anti-migraine medication prescriptions for executive staff at LSI Logic in the late 1990's. Despite aggressive programs to boost employee retention thru raises, promotions, stock options and greater flexibility in work assignments, the brain drain from LSI to startups and established companies in the new SoC sector was chronic, with Broadcom the most oft cited destination for departing employees. 

There was ample reason for it as well. Though LSI was a very strong company thru the latter half of the 90's, Broadcom was growing like a weed. It's product line of networking and broadband chips was strong and continually expanding as a result of the firm's incredibly aggressive M&A strategy. As a fabless semiconductor company, Broadcom also earned excellent margins on its products, fueling its acquisition spree. The company had a gigantic IPO in 1998, turning even the company's lab technicians into multimillionaires.

There were a slew of other similar companies founded in the 1990's that built sophisticated, highly integrated IC's and grew spectacularly - Marvell, Mediatek, Nvidia and others. What did these companies know that others did not?

The SoC Value Proposition

What is an SoC? Well, that's simple enough, I guess - it's a System on a Chip. But what is a System on a Chip?

Lots of companies like to call themselves SoC houses - it has a nice ring to it and tickles vacuous Wall Street types pink. Yet only a baker's dozen or so of chip firms can legitimately claim to have the demonstrable expertise to produce such products. And the expertise required is indeed a very steep barrier to entry in this segment. 

The Qualcomm Snapdragon 810 provides an illustrative example. The specifications for this smartphone applications processor are amazing - a 2 GHz 8-core ARMv8-A CPU cluster, 64b LPDDR4, LTE6/7, Bluetooth, full multimedia (audio, video, voice and graphics) and more, all packed together on a chip built in a 20nm process node. And that's just the hardware. Software support encompasses the entire stack, including applications such as HD voice & video, voice activation and others.

Obviously, SoCs don't exist only for the smartphone market. Broadcom cut its teeth on enterprise networking. Marvell started in mass storage, and was one of the pioneers in PRML. Ikanos built its company on DSL. Yet despite market segment differences, there are common themes between all of these companies: 
1. They have profound expertise in the development and integration of a tremendous variety of circuit functions - analog, mixed signal, DSP, CPU, embedded memory, GPU, accelerators and macros for a smorgasbord of industry standards and protocols in the three C's (computing, communications and consumer.)
2. Each company demonstrates profound system-level expertise in the market segments it serves. Thus, video decoder macros are designed to support not just industry standard specifications, but far more difficult 'stress' video streams that are considered the minimum acceptable quality level by broadcasters. The same holds true for the OFDM circuits of DSL or 802.11x designs, or blocks supporting various other standards - there is the know-how to 'over-engineer' the function so that it meets or exceeds the requirements of 2nd level customers (the end user customers of the OEMs to whom the SoC company directly sells.)
3. They capture the full software stack - from Firmware and OS up thru middleware to the application layer itself - and, bundled with GNU-based tool chains for the embedded processor elements, offer it as a product that can be further tailored by the customer for their specific system application. This was something that even in the early or mid 1990's was the exclusive domain of the OEMs themselves.
4. They have a deep understanding of the special features requirements - beyond just circuit functions - that are of value at the system level. A clear example of this is the many power control features common to smartphone application processors, such as adaptive voltage scaling, dynamic power switching, forward & reverse body biasing and so forth. 
5. Staffing reflects the importance of software over hardware, as there are usually 3x-4x more software developers than HW engineers. 

The depth and breadth of system and chip expertise necessary to execute on such designs is intimidating. The sophistication of these products is such that system OEMs orient their board designs around these devices, implicitly acknowledging the dominant position of the SoC in the system design. Accompanying this technical shift was a transfer of value down the technology chain, as evidenced by the revenue and margin strength of many SoC companies from the mid 90's to the 2008 crash. 

The Road Not Taken

One may appeal to genius, which is above all rules; which amounts to admitting that rules are not only made for idiots, but are idiotic in themselves. - Carl von Clausewitz

The entire SoC business and engineering framework violated multiple tenets of the operational canon for standard parts chip companies of the 70's, 80's and early 90's. Firms like National Semiconductor, Motorola, Cypress and Advanced Micro Devices built components with steadily increasing levels of circuit integration but were much more hesitant with regards to software, as they saw themselves as predominantly hardware companies. Any needs for significantly greater hardware integration were assumed to be the domain of ASIC houses.

SoC companies, by contrast, challenged the status quo and gambled on the idea that there was a huge market for much more heavily integrated chips that met 80% of the collective needs of a given market segment while providing a software distribution that would permit OEMs the necessary flexibility to tailor the chip to their particular needs. Their intuition proved prescient, as succinctly demonstrated by the collective success of the segment.

One can see from the foundational concepts of SoC companies what the sources are of their value-add to the market. The first is the personalization made possible thru a software distribution released with the SoC. The second is the extent of integration, both in terms of the hardware with all its disparate protocols and standards as well as the software with all its modules.

When you think about that second source of value-add in particular, there are some interesting parallels with what sophisticated IP licensing companies do. Take ARM or Imagination, for instance. These IP houses develop CPUs and GPUs for integration into chip designs. The individual chip companies that are their licensees effectively share the development and maintenance expenses of the complex IP between them and get access to it at much lower total cost than if they developed it in-house. In a sense, SoC companies scale this concept up dramatically by integrating multiple standard IP blocks, a software stack and a variety of system level features into a solution, allowing their OEM customers to share the development expense of capabilities they all require and receiving an SoC in return at much lower overall cost to themselves individually.

As in previous posts for this editorial series, the technical proficiency of SoC companies is captured in a 'heat map' presented below. There's no question that they are the best of the lot in the custom chip space.


Men's fortunes are on a wheel, which in its turning suffers not the same man to prosper for ever. - Herodotus

Though one could qualitatively state that the SoC segment is the strongest of the solutions on offer in the Custom Logic market, it is nevertheless not without its share of weaknesses. It can be broadly attributed to SoCs as an aphorism that they are able to cover around 80% of the system needs for a given application. Conversely, this implies that they generally fail to meet 20% or so of those requirements. The software distribution as 'universal balm' does not completely compensate for an SoC's shortcomings, since software-based supplementary functionality is inherently much slower and less efficient than its hardware equivalent. Finally, the fact that system designers need to develop their product around the limitations and constraints imposed by an SoC always grates on the design team to a greater or lesser extent.

There are now external factors that are undermining the value proposition of SoC solutions. Depending on who you ask, the demise of Moore's Law is either imminent or a de facto reality. In either case, the ability of SoC developers to continue adding features and stay on the value growth curve has been severely compromised. The growing frustration over a stymied Moore's Law and consequent hobbling of 3P improvement and feature innovation for competitive value-add was described succinctly by Karim Arabi, Qualcomm's VP of Engineering, at last month's DAC in San Francisco:

This squeeze on the ability to innovate is coming at the worst possible time. All semiconductor markets - including the segments served by SoC vendors - are either in decline, stagnating, or peaking, as has been discussed in depth in previous editorials. Such a powerful left-right combination of blows would stagger even the strongest company.

As a consequence, cost is beginning to dominate as a competitive factor and margins are dropping. This decline in profitability has already resulted in significant casualties. Over the last three years Broadcom divested from its TV, BluRay and Cellular Baseband groups. They are not alone in their growing financial distress - it's a widespread phenomenon in the sector.

The slowing and stagnation of SoC innovation is creating painful distortions in the foundational business model. In an effort to control and reduce costs, SoC houses are cutting staff and support for an increasing amount of their standard IP. The development and maintenance of cores for functions such as Ethernet PHYs and MACs, PCI Express, USB and other generic IP is no longer perceived by SoC firms as a value-add, but a sunk cost. Keeping engineering groups busy and up to date on such standard functions is now viewed as an increasingly unacceptable drain on the company's finances. Stated differently: the internal development and support of large standard IP portfolios is no longer considered a source of value or even a strategic necessity, but a trap.

Can SoC companies adjust to these new, difficult market circumstances and rebuild margins & growth momentum? Frankly I don't think so. The problem is not one of technology - the technical capabilities of SoC firms are awe-inspiring. The fundamental issue is one of organization.

No man can put a chain about the ankle of his fellow man without at last finding the other end fastened about his own neck. - Frederick Douglass

The very determinants of the SoC business model - skillful conformance to a wide selection of industry standards & protocols, aggressive pursuit of advanced process nodes on which to implement designs and driving schedules that conform to the product release rhythms of OEM clients - have resulted in neo-militaristic management styles that compel absolute conformance, along with project management dominated by the driving demands of specifications and schedules.

A significant proportion of SoC executives are convinced that they know how to squeeze the last drop of innovation and effort out of their employees. The industry is replete with stories along these lines - audio software developers, WiFi hardware engineers or video accelerator experts forced to work 120 hour workweeks, relentlessly driven for months on end, thru skin rashes, ulcers and other nervous maladies; HDTV or storage SoC hardware teams that have reliably fallen into lockstep with their OEM customer schedules, but 80% of whose internal communications consist of phrases such as "Yeah, well, that's what the block does, so deal with it", "This is YOUR problem now", "SHUT THE *$%# UP!" - and so forth.

Not all SoC companies are the same, and not every team at these firms has an atrocious working environment. Nonetheless, SoC vendors are very regimented organizations with linear thinking. The confluence of problems that these enterprises are facing now is, however, distinctly nonlinear. It is, in fact, a disruptive market event - a Black Swan.

Companies cannot react to Black Swans in the same manner as earlier problems. That is a recipe for failure. No SoC company is going to 'innovate' its way out of this. What is required now is the embracement of non-linear behavior and thinking.

It is incumbent upon SoC executive management to accept that their markets are being affected by macroeconomic forces which are different from those of previous slowdowns. The global market is - except for a very few isolated bright spots - still weak, with limited consumer disposable income, difficult employment environments and general corporate, finance sector & sovereign debt structures that are a drag on almost every national economy. These are major macroeconomic problems that will not resolve themselves in the next year or two, but will more likely take another 7-10 years to sort out.

Technology obstacles - the other 'wing' of the Black Swan - are just as daunting. Without a doubt, an enormous amount of effort over the next few years will go into further developing capabilities such as 2.5D and 3D IC, FD-SOI, finFETs and exotic memory architectures such as ReRAM to squeeze the final incremental benefits out of silicon. Even so, it is necessary for the SoC executive staff to accept that all these endeavors are the last hurrah for silicon in the same way that the Yamato and the Bismarck were for battleships in the face of the rise of naval aviation. They are stopgaps that may help to partially tide the industry over until carbon-based materials (e.g. graphene films, nanotubes and other technologies undergoing extensive R&D as discussed in previous articles) are ready to supplant silicon, but none of the new materials are likely to be production-ready for another 7-10 years.

In order for SoC companies to weather the storm of this intermediate period and emerge on the other end better prepared and stronger than their rivals, the determining factor will be how well these firms have managed to completely turn their organizational practices inside out. 

The intrinsic value of an SoC company is, of course, its people. Currently that talent is more or less chained down and frozen in place. That pool of expertise, initiative and genius is tapped only insofar as the top brass specifically permits and directs in what is a very top-down management system. To find clever and inspired ways of wresting increments of superior competitive value out of silicon & software, as well as prepare the enterprise to adapt itself to future technologies more quickly and ably than competitors, SoC companies must transfer authority down the chain of command to the worker bees and give them the power and support to take the initiative, feeding requirements and directives UPWARDS in the hierarchy. Without a great deal more freedom of action to do things and drive them, the enormous pool of skill and talent in the ranks will not feel comfortable taking initiative and responsibility for the company's future.

For those who scoff at such a 'preposterous and risible' idea, what follows is an example of how just this sort of approach has quite literally worked miracles.

Kelly Johnson and the Wizards of Burbank

As a lead designer for the P-38 Lightning II and the P/F-80, Kelly Johnson was one of the most brilliant minds in the history of aviation. He led the Skunk Works R&D division for Lockheed from 1958 to 1975 - a time where the group designed and successfully launched aircraft such as the U-2, SR-71 and F-117A. 

One day during the development stage of the SR-71, an urgent issue came up with regards to the plane's airframe. Kelly Johnson called the chief metallurgist down in one of the labs and instructed him to come up to Johnson's office to discuss the matter. The metallurgist, irritated at the interruption of his work, told Kelly Johnson point blank that he didn't have time for this nonsense and that if Johnson wanted to talk about it he would have to come downstairs - and hung up.

Kelly Johnson was the type of guy who told generals at the Pentagon that he was going to do things his way and if they didn't like it, they could go to Hell. The generals - not exactly wallflowers or shy & retiring types by any stretch - clammed up and gave way to Johnson. 

So what did Kelly Johnson do when his metallurgist gave him the brushoff? He bit his tongue, went downstairs to the man's lab and rationally discussed the issue at hand in order to come to a resolution.

Kelly Johnson put together a team of scientists and engineers that created aircraft 50 years ahead of their time. In effect, the Skunk Works turned aviation dreams into concrete reality. One of the guiding principles he followed that led to his astonishing success was to give full responsibility, authority and support to the people who were doing the actual work.

SoC firms play a leadership role in High Tech and have earned their prestigious standing thru great effort and sacrifice. Nonetheless, circumstances are changing profoundly, and to stay the course with the tried & true would be fatal to their future prospects.

So what does the future hold for the Custom Logic market as a whole? We'll discuss that next week in the concluding editorial to this series, as we once again embark on a journey to consult with the Pythia on the slopes of Mount Parnassus........

No comments:

Post a Comment

Feel free to comment or critique!