Is the Moore’s Law era finally at an end?
October 2016 | COVER STORY | SECTOR ANALYSIS
Financier Worldwide Magazine
Moore’s Law – that long-held golden rule of the electronics industry which states that “the number of transistors per square inch on integrated circuits doubles approximately every two years” – has been driving development within the technology sector since Intel co-founder Gordon Moore expressed his theory back in 1965. However, whether its principle continues to hold will have important repercussions for the sector, including the nature and direction of M&A activity.
Understanding Moore’s Law
Confusing the issue is the fact that there are at least two other definitions of Moore’s Law in circulation. The first is that the relative cost per component is minimised when you double the number of components every two years, and the second is that overall processing power for computers will double every two years.
Accuracy of definitions aside, Moore’s Law is not really a law at all. It is both an observation and an aspiration, according to Jack Ganssle, a consultant at Ganssle Group, having been a major springboard for innovation throughout the tech sector for many years. The electronics industry has worked very hard to maintain the doubling rate observed by Moore all those years ago. Intel’s tick-tock strategy – which Mr Ganssle characterises as improving the design of their microprocessors in year one to make them faster, then in year two shrinking them to double transistor density – is indicative of this trait.
The end may be nigh
However, for all its influence, the Moore’s Law era is considered by many to be approaching its end. Some, such as Intel’s former chief architect Robert P. Colwell, have even predicted when this end will come. Colwell stated in 2013 that 2022 will be the historic year.
Others are not convinced. “Moore’s Law was going to end when transistor geometry hit 1000 nm (nanometers), then at 700 nm, then 450 nm,” says Mr Ganssle, “but today we’re at 14 nm. At 40 nm it was predicted dead because the gate oxide – a critical insulating part of a transistor – had to be 1.2 nm thick, which is the diameter of 5 silicon atoms. Guess what? New kinds of oxide were invented so we could keep shrinking transistors without making the gate oxide thinner. It was dead at 20 nm because of another problem with gates. Instead, Intel invented the tri-FET, a three dimensional transistor.” To date, innovation always seems to find a way.
Whether the consensus is that Moore’s Law is developing, dormant, doomed or dead, what is beyond doubt is that if not a definite abandonment, we are witnessing a considerable slowdown at present. The impact of this on the overall tech sector and its M&A activities are likely to be hugely significant in the years to come – particularly as tech dealmaking has experienced a boom in recent years, led by SoftBank’s £24bn offer for ARM Holdings in July 2016.
Suggestions that the long-established Moore’s Law is now at a point of being put out to pasture are nothing new. Indeed, many an industry practitioner has speculated as to how an abandonment of Moore’s Law, by the semi-conductor industry for example, is likely to impact the tech space as a whole.
“For businesses that have built their brand on doubling CPU performance every two years on the bleeding edge of manufacturing processes, the writing has been on the wall for many years,” says Derek McAuley, professor of digital economy at the University of Nottingham. “However, in the office computing environment, the performance of desktop computers surpassed the needs of most office works about 10 years ago; the turmoil in this market long predates the Moore’s Law doomsaying.
“Likewise in the server space, with the emergence of cloud services, scale out in number of customers served is critical, not peak single processor performance, and most of the large cloud suppliers have already moved to using larger numbers of lower performance and lower power processors as the most economic solution,” he adds.
According to Apek Mulay, a business and technology consultant at Mulay’s Consultancy Services, the end of the Moore’s Law era has been due to poor return on investment (ROI) resulting from the increased cost of future investments and lower consumer purchasing power in the economy. “While the semiconductor industry has tried to address the problem of increased cost of investment through M&A in the tech space, it has failed to boost consumer purchasing power in the economy. The real job creators in an economy are not only the producers but also the consumers. Hence, for Moore’s Law to continue into the foreseeable future, there has to be a growth in the consumer purchasing power in proportion to the exponential growth in the number of transistors on an integrated circuit, or for macroeconomic reforms to be undertaken toward a balanced economy to avoid huge unemployment during economic downturns,” he says.
For Mr Colwell, there has been no ‘abandonment’, apparent or otherwise, of Moore’s Law. “The electronics industry is just as addicted to the endless bounty of Moore’s Law as it has ever been,” he says. “What other technology in all of human history has exponentially and simultaneously improved its performance and cost metrics over five decades?” He adds that the industry will “ride the Moore’s Law horse until it dies out from under them” for the simple reasons that there is no alternative available to them. “I believe that day is coming soon,” he notes sagely.
Supporting the contention that Moore’s Law will continue until it is eventually superseded by another significant industry development, is the related phenomenon called Dennard Scaling, or MOSFET scaling. This scaling law, which some call the ‘quieter partner’ of Moore’s Law, originated in a 1974 paper co-authored by Robert H. Dennard.
In summary, Dennard Scaling states that “as transistors became smaller, their power density remained constant — meaning that smaller transistors required less voltage and lower current”. Therefore, while Moore’s Law states that every two years or so we would be able to pack a greater number of transistors into the same-sized area, Dennard Scaling provided the assurance that the same transistors would be cooler as well as needing to draw less power.
“Moore’s Law isn’t just about doubling the number of transistors,” points out Mr Colwell. “In its heyday of the 1980s, Dennard Scaling was also in effect, where the power supply voltage for transistors was lowered with each new complementary metal-oxide semiconductor (CMOS) generation. Because overall power dissipation was a function of the square of the supply voltage, those new transistors were free and the new chip was not noticeably hotter than the old one.”
However, Dennard Scaling came to an end around 2005 – a demise which, according to William D. Gropp, Thomas M. Siebel chair in computer science at the University of Illinois, was a result of it ignoring the ‘leakage current’ and ‘threshold voltage’, which establish a baseline of power per transistor – in other words, the chips overheated.
Post-Dennard Scaling, Moore’s Law of course continued, but now with its longevity under almost constant scrutiny. “We in the industry have known all along that eventually some physics-related problem would appear that would fundamentally constrain Moore’s Law,” admits Mr Colwell. “In fact, we thought we had hit such limits a few times already, such as when the silicon feature sizes started to be smaller than the wavelength of the light sources used to create them. Engineering cleverness found ways around these ‘false alarms’ in the past, encouraging some to believe today’s challenges will also be faced down.”
Since Moore’s Law has driven development in the tech sector for over five decades, it is likely to retain a certain cachet for some time to come. And yet, the ability to adapt and evolve is a key requirement in any environment, and the tech sector is no exception. So, if Moore’s Law is indeed on a downward spiral, perhaps irrevocably, how is the tech sector as a whole to respond?
It may not be long before the industry has to come to terms with the end of the Law and all it stands for, and then somehow find new ways to foster innovation, drive growth and encourage M&A investment.
“Some areas of the tech sector already display the creativity to adapt,” vouches Professor McAuley. “They have used the transistor bounty offered by Moore’s Law to develop deeper integration of diverse components, building single chip systems that are the core of modern consumer equipment and greatly expanding the market for electronics. This massive growth is affecting the economics of component cost at the core of Moore’s original paper. There also continues to be scope for continued creativity in the use of all those transistors already available.”
Others believe Moore’s Law, or something very like it, has to continue due to growing customer demand for ever-increasing capabilities. This demand will push tech sector evolution and adaptation into the future. “You would be astonished at the amount of hardware in the system-on-a-chip, which is the biggest single chip in your phone,” says Mr Ganssle. “Some suggest just giving up on process shrinks and instead putting more integrated circuits on a circuit board. But that has problems because these high-speed signals simply cannot propagate over the long distances – think an inch or two – on a circuit board. Rather, they have to stay inside the chip. Today, there is all sorts of innovative 3D technology coming along, such as V-NAND and the 2.5D technology developed by Xilinx, which is extremely interesting.”
Impact on M&A activity
If the Moore’s Law era is indeed drawing to a close, it will certainly have an impact on mergers & acquisitions (M&A) activity in the tech industry going forward. “M&A activity will be driven by the sector’s creativity in the use of the transistor bounty that Moore’s Law has already provided,” suggests Professor McAuley. “I expect to see M&A activity around companies that can produce hard design intellectual property (IP) which uses field-programmable gate array (FPGA) elements, but especially around key IP that could be converted into fully custom coprocessors.”
However, in the view of Mr Mulay, free market or competitive capitalism is critical if Moore's Law is to continue for the foreseeable future. “The US economy has been transformed from free market capitalism to monopoly capitalism due to increasing M&A happening in every sector, including the tech sector,” he argues. “This process has resulted in a growing gap between wages and productivity, resulting in excess supply and poor economic demand bringing Moore’s Law to a premature demise.”
Despite the theories about its imminent demise, there are a number of Moore’s Law principles which industry practitioners believe will continue to have resonance, even within today’s highly complex technological environment. This influence is set to endure whether the Law is approaching its endgame or not.
Mr Mulay envisages a future that depends on the global semiconductor industry undertaking free market economic reforms to ensure that wages keep pace with productivity. “The progress of Moore’s Law has essentially been the progress of only supply side economics,” he suggests. “However, as long as free market reforms can ensure that economic demand grows in proportion to the growing supply of consumer electronics, the continued progress of Moore’s Law can continue to benefit today’s knowledge based economy.”
To be sure, minimising the cost per component will remain part of the focus. According to Professor McAuley, industry practitioners should be looking beyond materials science and physics. “Chip design these days is very expensive business,” he says. “Thirty years ago a small group of post docs and PhD students could design a chip with tools they made themselves. The current tool chains and the complexity of chip design need significant work to reduce cost, as do the tools for integration of components into systems and systems into networks. There is still plenty of scope for technology to drive down ‘per component’ costs; however, it will be less about single chips and more on total system design.”
The mantra ‘better, faster, cheaper’ summarises each new generation of chips that Moore’s Law made possible – although the addition of ‘sooner’ may also be necessary. “There are certain aspects of the computer industry’s history that relate to Moore’s Law that have pervaded the thinking of all concerned,” says Mr Colwell. “I think the computer industry will perform amazing technology feats, marketing sleight-of-hand and legitimate product improvements that do not require the Moore’s Law underpinning, to keep the industry running – or at least appearing to be running – for at least a decade post Moore’s Law.”
It is important to take a step back, however, and accept that the computer industry should not be regarded in isolation, as an industry independent and distinct from other industries such as communications, transportation, science, entertainment and medicine. The nature of our interconnected modern world has important implications for the development of computing – and vice versa.
“Computers are now as necessary to future growth in many industries as electrical power and stable governments,” suggests Mr Colwell. He cites a phenomenon over the past 50 years that has seen the repeated emergence of new computing technology, immediately enable new findings in different areas. In science, it led to the human genome, protein folding and drug discovery. It has facilitated better communications through the internet, social media and video compression technology. “One of humanity’s upcoming challenges is to keep this momentum in the face of a computer industry that is heading for a fast decline,” he concludes.
For those experts who believe we are approaching, or have already witnessed, the final throes of Moore’s Law, the excitement lies in anticipating what new innovation or set of principles will rise to take its place.
© Financier Worldwide