Just as Moore’s Law hits its long-awaited “wall” and runs out of steam, it has intersected with Metcalfe’s Law and combined to put both on a new trajectory toward unleashing what we now vaporously call the Internet of Things (IoT).
But make no mistake, it is really about the Internet of Us (IoU), and as it explodes in the next five years, it will be a boon to a maturing semiconductor industry facing process-node stagnation, and a life-changing experience for everyone else.
Fifty years ago, when Gordon Moore first made the now famous observation that ICs would double in density roughly every two years for a given minimum cost, he could not have imagined that he would end up being a proof point for some quantum theorists’ stance that by simply observing something, one can change the outcome (Schrödinger's Cat). But he surely did, albeit indirectly.
(Read "Moore's Law at 50.")
He was not to know, of course, that the observation would be adopted as the International Technology Roadmap for Semiconductors (ITRS), which turned his observation into fact, and the rest is history—from dramatic falls in the cost of transistors, from $5.52 in 1954 to one billionth of a dollar in 2014, to increased functionality per milliwatt.
This created a virtuous cycle, whereby transistor scaling brought better performance at ever-lower cost, which brought about exponential industry growth as the applications for semiconductors grew. This allowed further investments in new technology to continue scaling ever downward, past 0.18nm to 0.13nm, then 90nm, 65nm and now 28nm, 20nm, 14nm and below.
However, in a keynote presentation at the ISS 2015 gathering in Half Moon Bay in January, Scott McGregor, president and CEO at Broadcom, described how the cost of the manufacturing technology and processes required at 28nm and below have created a situation whereby the cost per transistor has started to rise, for the first time in the industry’s history.
The analysis and the conclusion were first postulated by Dr. Handel Jones, chairman and CEO at IBS Inc. in a white paper, “Why Migration to 20nm bulk CMOS and 16/14nm FinFETs is Not Best Approach for Semiconductor Industry”, published in January 2014. Handel continues to update the exact numbers, but the conclusion emphasized by McGregor at ISS was that 28nm may now become the optimum-cost node.
For sure, manufacturers of high-margin devices such as network processors and server chips and high-density memory can justify the cost of moving below 28nm today, but for everyday silicon, 28nm will be the sweet spot.
Not All About the Chip
So now what? McGregor’s answer is to do more with the silicon and processes we have, through better design engineering versus process engineering. In Broadcom’s parlance, this translates to 6GHz digital RF sampling, 100 Gbit/s mm-wave (1 THz) connectivity (1 THz range) and 100 Gbps broadband access speeds, and much more, all by 2030.
That may all come true, but one slide stood out, “It’s all about the chip.” That is not entirely true.
Semiconductors are an enabler but no longer with silicon as the be-all and end-all of high tech. Software is increasing in importance, as more time and human capital are spent developing firmware and accompanying software than on the ICs themselves.
Another important trend was happening at that time too. Wireless networking was entering everyday vernacular—from Wi-Fi to Bluetooth and ZigBee—but it was all disconnected, expensive and hard to use. We needed a “unification theory” for wireless and wired communications. A way to connect the dots. We looked, but nothing showed up. Protocols and interfaces remained apart. Still, embedded and consumer device developers fought on and continued paving the path to machine-to-machine communications.
Then, at the beginning of the new millenium, companies such as Facebook, Google, LinkedIn, Twitter and now Snapchat and many others, became the new face of high tech. This is an important inflection point as it points to a deeper need than one-way video downloads, faster gaming, graphics or faster download speeds for work-related files. It pointed to a need for people to connect, symmetrically and in real time. And connections were happening, quickly.
Then Came MetCalfe’s Law
All this resonated with another law—Metcalfe’s Law. This states that the value of a network increases exponentially with the number of compatible users. Attributed to Robert (Bob) Metcalfe, the father of Ethernet, the law was first named as such by George Gilder in 1993. The key words in the law are "value", "exponential" and "compatible."
What Metcalfe’s Law tries to do is put a value on a network as more users get added. Some have debated the value as represented by the law, but the principle holds. As more users get added, a network increases rapidly in value. In the past 10 years, that network has become the Internet, the unification point, and the number of connected users of services of many types—from search to Snapchat—has gone exponential.
A key factor in the Metcalfe equation is the value of those connections. Direct person-to-person communications have been the over-arching goal with Internet services, and that is given high value by stockholders. Thanks to Moore’s Law, the cost of making those people connections continues to fall. So does the cost of connecting "things", from remote sensors in the Himalayas to light bulbs in the basement. In fact, the cost is so marginal and the integration so well streamlined that it is almost odd to not IoT-enable everything. Just ask Freescale, Silicon Labs, Texas Instruments or Microchip.
At the same time, the value that can be extracted from billions of these sensors, thanks to Moore’s Law’s enablement of low-cost, high-performance data analysis has also gone exponential as businesses adapt quickly to data-based decision making enabled in large part by entities such as Intel’s Internet of Things Solutions Alliance. The Alliance, now comprising over 250 members, is dedicated to providing the hardware, software, firmware, tools, systems integration and end-to-end analytics needed to make IoT work for maximum return on development investment and time. Those are high-value connections.
Of course, Intel also wants to sell chips. Lots of chips. And like many other companies, it has developed kits to ease integration, as has Broadcom with its WICED line of connectivity kits. The list goes on through a who’s who of the semiconductor industry.
There is a good reason to be optimistic that IoT will drive revenues. According to Bill Morelli, director of IoT, M2M and connectivity at IHS , the installed base of IoT nodes was estimated to be 8.6 billion at the end of 2014, and is projected to increase to 50.7 billion by 2025.
While at first glance, these may seem to be low-margin ICs for sensor signal conditioning, data conversion, signal processing and wired and wireless communications, in fact billions of devices are feeding zetabytes of data upstream to servers via network processors. So the virtuous cycle may well continue past 28nm, but the low-leakage requirements for battery-driven applications will keep connectivity ICs for IoT somewhere between 90nm and 28nm for some time to come.
It’s Not About ICs, or Things
While McGregor’s point that ICs were a critical part in enabling the next generation of innovation around the 28nm sweet spot is valid, it is clearly not all about semiconductors, and maybe he did not really mean to imply that. Nor is it about "things." It is really all about ‘us’ and how we can connect things to make them work for us, to make our lives easier, more fun, or our businesses more profitable.
Since the first stories were carried from one generation to the next, to the first drawings and writings, to the first books, email and Facebook connection, it is really all about our need to connect across space and time. The Internet of Us (IoU) will accomplish that and William Gibson will be proud. As will the Borg.
What Moore’s Law has enabled, Metcalfe’s Law will ensure exponential continuity. Of course, one has to give a nod to Shannon and Nielsen, but that is another story.
Questions or comments on this story? Contact email@example.com