Chips Ride Atop the Leader Board, Part 2
Macom’s Robust Lineup of Chips for Data Centers and Cellular Network Base Stations
volume xvii issue 3
Macom, Inc. of Lowell, Mass. (NASDAQ:MTSI) has been busy building semiconductors for more than sixty years, primarily microwave applications for the military. Wall Street analysts consider Macom the last remaining pure play focused on microwave chips for radio. These chips can be used in all sorts of applications. Commercial cellular telephone frequencies in North America occupy spectrum at about 700-900 MHz for 3G/LTE, and microwave frequencies north of 2 GHz employed with 4G/LTE and in planning for 5G/LTE networks.
Today we categorize chip manufacturers based on whether they fabricate their own – a foundry with etching and photolithography equipment. Otherwise we call these fabless, which means the companies design chips but outsource the manufacturing. Macom considers themselves ‘fab-lite’. They maintain a facility near headquarters outside of Boston for high-end applications, testing, and R&D. The bulk of the manufacturing business is outsourced to six foundries in Asia.
CEO John Croteau likes to proclaim that Macom’s future is blessed with the insatiable demand for communication in all varieties, and has centralized Macom’s strategy on chips placed with OEM’s building Ethernet port cards and switches for data centers. He points out, that of bits traveling in the cloud, three quarters of the traffic is inside the data centers. Macom estimates there are 40M Ethernet ports in North American data centers today, and they are using a planning number of 70M by 2020.
Macom’s big deal right now is a growing foray into optics. Via acquisitions for the most part, they build chips that produce pulses of light in a single wavelength, and the chips needed to amplify the light connected to fiber optic cable. Macom completed an acquisition of Applied Micro Circuits Corporation (NASDAQ:AMCC) in January following a trend of consolidation in the business that has been going on for a couple of years. Macom paid $770M in cash and stock for the company.
The winning piece with the acquisition is Applied Micro’s progress with PAM-4. In a layman’s description, PAM-4 is a set of rules that describes how to translate light over a fiber optic link into usable information. Data centers have endorsed PAM-4 as the optics standard for moving to 100 Gbps and faster speeds. This is a not-so-often case where less is more. Important because PAM-4 reduces the lasers and drivers needed in current circuit modules from four to one. Multimode fiber has some speed limitations above 100 Gbps that are avoided when using a laser source transmitting only one wavelength over the fiber.
Speed with this architecture is now a function of grouping modules together rather than multiplexing. All you need to know is that modules using only one flavor of light are faster and cost less to make than multiple wavelength solutions. Four 100 Gbps PAM-4 modules can be bound together to make a 400 Gbps ensemble on one Ethernet port. The data center business predicts the rollout to fully 400 Gbps capable hardware by 2018/2019, just around the corner.
The other jewel with the AMCC acquisition brings hardware-based encryption via the IEEE 802.1AE link-layer standard. This standard defines how an Ethernet card keeps track of an encrypted data signal coming and going between two capable ports. 802.1AE is called MACsec. Cisco has a significant role along with other IT communications hardware companies in designing the 802.1AE standard. Prior to the merger, AMCC was already way out in front of the pack as a strategic supplier of MACsec technology to major OEMs. Without the hardware encryption, data centers would be easy cyber-targets for malicious hackers at the network layer.
Together, MACsec (hardware encryption) and PAM-4 (optics-based modulation) technologies acquired through M&A have given Macom a seat at the table for architectural discussions with the world's leading enterprise and cloud data center providers to deliver single-mode, fiber-based optical transmission. PAM-4, now adopted by the IEEE as a commercially suitable optics solution, is expected to be the most cost-effective and efficient enabler of 100/400 Gbps transmission in data centers for years to come.
Last September in Düsseldorf, Macom and BrPhotonics made a demonstration at a trade show with AMCC’s 100 Gbps PAM-4 technology. Macom successfully repeated the demonstration with Sumitomo Electric, at the Optical Fiber Communications Conference – 2017 in Los Angeles, and announced the new chips as soon to become commercially available.
The other really big deal in Macom’s portfolio is the progress they have made replacing Gallium Arsenide with Gallium Nitride for microwave power amplifiers. And at reasonable cost with non-government applications. New GaN amplifiers will be installed in the base stations for cellular telephone antennas at the tower. The most lucrative, near-term opportunity lies with the rollout of 4G/LTE base stations – a billion-dollar semiconductor market – where GaN plays a critical role in the ability for carriers to expand network coverage, increase data rates and reduce energy costs with operating base stations.
Cell towers and base stations never sleep. The networks operate 24/7, 365 days a year. The more robust the amplifier equipment, the better, and major carriers like AT&T and Verizon can afford switching to GaN. These benefits are rooted in the material science of GaN as a semiconductor, which yields superior power efficiency, signal bandwidth and clean operating frequencies for the most advanced 4G/LTE and future 5G/LTE networks.
There is a bit more relevant detail with the financing that surrounds all of this M&A activity, and Macom issued their IPO in 2012. The stock is closely held by founders. Stay tuned for a follow-up discussion in this blog about the financials, to be posted in July.
Copyright © 2017 New Edge Analytics, All rights reserved.
Chips Ride Atop the Leader Board, Part 1
Layering GPU’s and CPU’s to Ramp Up Processing Speeds in Data Centers
volume xvii issue 2
Semiconductor manufacturing is the most complex industrial feat we take on as a species - worldwide. Modern microprocessors are incredibly complex housing more than a billion transistors, each about .02% the width of a human hair.
Semi (chip) manufacturing seemed until a few years ago, to have matured, enjoying plenty of worldwide building capacity despite the investment and technology challenges. But then the industry suffered brutal price competition for central processing units (CPU's) and memory chips as the business commoditized.
Due to the continuing evolution of the Internet and the ability to merge software with chips, manufacturers again have the ability to create economies of specialization, allowing suppliers to set prices in certain markets.
Look at a company like Intel whose stock price showed a secular bear market pattern - lower highs followed by lower lows - for a decade. Intel suffered the tech bust in 2000 and it was 2009 until it seemed to turn a corner. An extended period of low interest rates certainly goosed capital and R&D spending. Intel both designs and builds chips. The foundry business is extremely capital intensive.
Now we have a growing list of ‘fabless’ semiconductor companies who design the chips and outsource some or all the manufacturing to the foundries. I have long held the belief that the day would come when semiconductor companies could cheaply burn applications onto chips, shorten the manufacturing cycle, and better control inventory pushed into the supply chain. And that's exactly what's happening.
Nvidia (NVDA:NASDAQ) is the Wall Street darling in the space of bit-level processors, up 600% in five years, 200% last year alone. Remember when Nvidia was the graphics card you bought for your PC with a special chip that made the monitor look great without slowing it down? Pre Dell, XBox, eBay and Amazon, you bought the card and the games from ComputerLand, Radio Shack or Best Buy, wherever you would go to buy a PC. You controlled the game from your keyboard, maybe you bought a joy stick.
Guys who worked for me when I ran a help desk at a large financial firm in 1995, figured out how to group-game on the Ethernet behind our firewall. Our workstations were video enabled for bit-mapped graphics. They thought that I didn't know, but I allowed it and the network administrators never caught on. Now these chips are capable of heavy duty bit flipping for lots of high performance computing applications. No longer a chip on an insert-able card, it is called a Graphics Processing Unit (GPU) and can run alongside the CPU as a dual processor to make servers run faster.
Nvidia has transformed from a special-purpose semiconductor company for games and graphics, to one that can build general-purpose processors, like a CPU. NVDA would rather call their solutions a platform, offering a suite of chip-enabled applications. The marketing slant is in their pubic presentations and financial reports.
The thing that has really changed our company, what really defines how our company goes to market today, is really the platform approach, that instead of just building a chip that is industry standard, we created software stacks on top of it to serve vertical markets that we believe will be exciting long term that we can serve. And we find ourselves incredibly well positioned now in gaming, in AI and in self-driving cars.
NVidia CEO Jen-Hsun Huang
Bank of America Merrill Lynch (BoAML) published a Semiconductor Playbook in January and it is convincing that NVDA has leadership - as in top of the heap - with these chip applications: PC gaming (still 60% of NVDA revenue with an 18% CAGR), Artificial Intelligence (deep learning), AI for self-driving and co-piloted cars, and a budding Virtual Reality (VR) business.
Beyond the predictable cash flow from the gaming business, if just one of these new businesses blooms and NVDA sticks to a fabless model, they win. Regardless of competition that validates their markets, they are first to the dance with a meaningful customer base and financial commitment, and negligible CapEx.
Artificial Intelligence is the biggest wager in bit-level computing per Goldman Sachs, who sees the current addressable market at $5B - $10B in annual sales. Compare these estimates to total annual sales in the semiconductor industry at $350B. That is a lot of headroom. As of the third calendar quarter 2016, Goldman Sachs estimates there are about 1500 AI startups worldwide, nearly all running the applications in the cloud. AWS, Azure, and IBM Cloud are favorites to house co-processors with burned-in apps, designed to easily scale along with CPUs in data centers. Startups fuel the bleeding edge innovation, venture capital, private equity investment and M&A pipeline.
Copyright © 2017 New Edge Analytics, All rights reserved.
Hello World!
Qualcomm’s Merger With NXP Semiconductor and the Interet of Things
volume xvi issue 10
These blogs and those to come express my thoughts about news in the IT industry impacting meaningful long term trends - microsteps in the evolution of IT that cannot be easily undone. Newsworthy this month is the merger announced between NXP Semiconductors and Qualcomm. Originally pegged at $30B, QCOM proposes to acquire NXPI per news late last week for $110 per share, an all-cash deal. Shares of QCOM are rich at this point and the prevailing advice is to assume a risk-off position here; the deal could fall through, corporate earnings could disappoint, the election is a wild card for stocks anyway you look at it. If the acquisition is delayed due to regulatory action, then NXPI is a good short here if you have risk appetite.
Valuation discussions are ongoing at this writing. Both companies have, for me, a most compelling story in the future of the Internet of Things. NXPI – originally the semiconductor manufacturing arm of Koninklijke Philips N.V – only last December acquired Freescale Semiconductor for nearly $12B. And Freescale has pedigree as the original semiconductor business at Motorola, added to their portfolio in 1948. Motorola was one of the first technology companies to make radios for the military, and in recent decades, many applications with semiconductors in the automotive industry. Think chip-enabled safety and security features, and the Google Self-Driving Car.
Motorola spun off Freescale in 2004. When NXPI and Freescale combined, they at once became the largest manufacturer of chips for cars. NXPI also has the lion’s share of the market for near field communication chips that enable tap-to-pay features used in mobile phones. This technology will increasingly show up in the guts of hardware that allow IoT devices to communicate. Not to mention, improved capabilities of smartphones, ID chips for credit cards and the like.
Qualcomm of course was a pioneer with spread spectrum technology (CDMA modulation) that became the commercialized wideband radio access method for wireless phones in the 1980’s. Spread spectrum technology is the foundation of modern 3g and 4g cellular networks. Qualcomm’s OmniTRACS system for truckers was one of the first of its kind used as a commercialized CDMA application.
Inside the Beltway this merger is a sentimental win, as all of these technologies grew out of military research, government and private partnership laboratories, that were later enabled by companies who made the right choices along the way in taking advantage of disruptive improvements in radio – further enabled by the Internet.
Other risk-off comments: Small cap and mid cap stocks are overvalued at current prices and should be avoided. If the market experiences a sharp decline or consolidation, large and mega-cap stocks with solid earnings growth and healthy dividends would be good portfolio additions. I typically look at a stock’s current price-earnings multiple and evaluate as over or under its average multiple for the prior decade, and buy typically when the ratio is under its long term average if nothing has changed with the stock’s fundamentals. Dividend yields among large cap stocks are fairly valued.
Tom Finkenbinder
Copyright © 2016 New Edge Analytics, All rights reserved.