Sunday, May 4, 2014

High Tech "State of the Union", By the Numbers - Part 3: The Stone Masons

With chiseled touch
The stone unhewn and cold
Becomes a living mould,
The more the marble wastes,
The more the statue grows. - Michelangelo

In this chapter, we'll look at three leading companies that represent what Silicon Valley is traditionally all about: semiconductors. It is from the base of intricately worked silicon that all creativity and innovation in High Technology springs - PCs, servers, blades, home gateways, supercomputing, smartphones, edge routers, LTE base stations, USB Flash memory sticks, HDTV, Social Media, eCommerce, the Cloud and on & on. If there are to be any future revolutions in High Tech, they will start from something invented here - at the chip level.

Qualcomm holds a dominant position in smartphone application processors with a 54% market share, while Apple and Mediatek trail far behind at 16% and 10%, respectively, in this $18B market (source: Strategy Analytics Q1 2014.) Broadcom develops SoCs for broadband, wireless and consumer electronics applications, and though not the leader in every market in which they participate, Broadcom commands a strong presence across all Communications and Consumer segments. Xilinx, the perennial king of programmable logic, counts enterprise networking as its mainstay, but also participates to at least some extent in the wireless, consumer, automotive, industrial, high performance computing and military/aerospace segments. 

These three companies continue the four decade history of High Technology firms with an extraordinary collection of technical talent drawn from across the globe trying to "carve value out of rock." Together, they provide an excellent barometer of all semiconductor markets other than personal computing & sensors and offer an indication of the underlying health of systems companies in communications, consumer electronics, industrial/scientific/medical, automotive, datacenter and defense sectors. 

The revenue numbers below are in $B, starting in calendar Q1 2008 to the present.

Qualcomm - This company has long been a market powerhouse in wireless communications, particularly in mobile phones, and its revenue growth shows it. The last four years in particular have been happy ones for Qualcomm as its fortunes followed the spectacular growth of the smartphone market. 

Nevertheless, a price war for both phones and services erupted during 2013 and both Samsung and Apple are forecasting lower smartphone growth for 2014. One can see the effect on Qualcomm's revenues since Q4 2012, as growth tapered and, if Q1 2014 indicates the start of a trend, may have begun to reverse. Clearly the smartphone market appears to be approaching saturation, which suggests Qualcomm's investors will begin to place mounting pressure on executive management to refocus the company and transform it from a "one trick pony" into a more broad-based Technology vendor.

Broadcom - It could be fairly argued that the SoC was invented by Broadcom, to the pain and regret of many an ASIC house. By developing chips that comprised at least 80% of market requirements for a given segment and leaving systems houses the opportunity to personalize the offering thru a bundled software stack and tools, Broadcom allowed system vendors to reduce fixed overhead costs and TTM - a one-two combination punch that no ASIC supplier could hope to match.

Yet despite its broad reach in Communications and Consumer markets, along with an incredibly rich IP portfolio, Broadcom has been essentially flat since the end of 2010. This would suggest extended softness in its markets, all of which ultimately depend on the individual consumer. Because of the continual paucity of consumer discretionary income, electronic appliances and gadgets have seen weakening sales, the effect of which has trickled down to both wireless and wired connectivity. Thus, Broadcom is now effectively trapped in stagnant markets.

It would behoove Broadcom's management to begin investing its extra cash in either internal applied R&D or M&A actions to expand its market reach. If Broadcom does not find a way to expand into sensors, analog or brand new markets based on non-silicon materials, the company will follow in the footsteps of Intel down the road to oblivion.

Xilinx - Looking at it fairly and objectively, Xilinx's revenue performance is pathetic. The company's strength in wireline communications should have translated into healthy growth as datacenters and enterprise routers upgraded to support 10Gb and then 40/100Gb bandwidths. Yet it appears that Xilinx revenue hasn't even kept up with the level of inflation. 

The problem is one of technology. Xilinx has integrated multipliers, high bandwidth serial interfaces, memory, hardened Ethernet MAC, Interlaken and PCIe cores and various other higher level embedded hardware features into its programmable arrays. The company has not been neglectful on the software front either, with ported applications, OS support, API frameworks and middleware developed independently or thru third party partnerships. Combined with advanced work in packaging (3D-IC) and development tools, one would think that Xilinx has done all it can to earn the description of "programmable SoC vendor."

To be sure, in every succeeding product generation Xilinx has made enhancements to clocking, routing and CLB resources while also expanding its embedded features. Nevertheless, the heart of these bit-level processors has not changed all that much. 

FPGA architecture has evolved - but only incrementally. Any FPGA designer familiar with the original XC2000 series would find the logic cells of the Virtex-7 to be tantalizingly familiar. Each successive product family includes several additional bells & whistles surrounded, as always, by a sea of LUTs. As a result, mainstream FPGAs are still slow, inefficient, power-hungry and expensive devices and remain fundamentally uncompetitive with ASICs and SoCs for the genuinely high volume opportunities.

Yet Xilinx continues to adhere to its 30 year old mania to convert all of electronics to a reprogrammable paradigm. Rather than continuing to indulge in this fantasy, Xilinx needs to confront the realities of High Tech - that all three primary semiconductor markets (the three C's of communications, computing and consumer) are stagnant, secondary markets (mil/aero, medical, industrial, automotive, datacenter/high performance computing) are either stagnant as well or declining, and both consumer and enterprise budgets are deeply distressed, seeking not only rock bottom pricing but also a high level of genuine, demonstrable value.

Certainly there is room for creativity in the programmable logic market. The runaway success of Lattice thru the acquisition of Silicon Blue and its ultra low cost FPGA line, as well as Tabula's CPU - like OTF (on the fly) configurability and its attendant density, cost and performance benefits suggest that there are many, many ways to 'skin the cat' differently in programmable logic.

Nonetheless, Xilinx doesn't necessarily need to drastically redesign its programmable fabric. It does, however, need to rethink the utility and applicability of its foundational technology. Xilinx needs to begin treating reprogrammability less like some canon of religious orthodoxy and more like a prime technical capability, but only one of many 'pieces of the puzzle' in its chips. 

Granted, Xilinx invented the FPGA. Fine. But get over yourselves, fellas! I'm sure RCA was extremely proud of its vacuum tube division back in the 1950's.

The design emphasis going forward should be on hardened CPUs, DSPs, fixed function standard cores and accelerators, with programmable fabric implemented only in the portions of a design where reconfigurability can add clear value. There are many obvious benefits to bit-level programming in carefully defined portions of an SoC design - the external memory interface and control logic, variable protocol support, internal bus monitoring for QoS, multimedia post processing and so forth.

With its 30 years of programmable logic applications experience, it should be child's play for Xilinx to identify many other applications where a programmable fabric in the role of an embedded core can add value as a competitive differentiator at the system level , while the rest of the design is implemented with fixed functionality optimized for the 3P's (price, performance and power.) Such an approach would create an opportunity for Xilinx to grab significant share at the expense of SoC vendors and put the company back on a path of healthy revenue growth.

In the next chapter, we will look at our last set of High Tech firms - the ones whose defining purpose is to invent the future.

No comments:

Post a Comment

Feel free to comment or critique!