Today's semiconductors consume less and less power but it takes more and more to make them
The electronics industry has a strange, often tense relationship with the worlds of energy efficiency and all things 'green'. It wants to be seen as a great enabler of cutting carbon emissions and better husbandry of the world's resources.
Intel, the world's largest chipmaker, has made this lofty claim: 'The [initial] conversion to the energy-efficient Intel Core microarchitecture saved up to 26TWh of electricity between 2006 and 2009 compared to the technology it replaced – equivalent to eliminating the carbon dioxide emissions associated with the annual electricity use of more than two million US homes.'
Beyond that, semiconductor companies generally remind us that they are extending the battery life and functionality of all manner of mobile devices, and that even their most powerful processors are today designed specifically to reduce the massive amounts of energy needed simply to cool today's ubiquitous server farms.
Then there is electronics' role in driving down the cost of solar energy, specifically by bringing Moore's Law to bear on photovoltaic technology and pushing it towards the crude acquisition cost target of $1 per watt (today's purchase price stands at around $4.50).
In contrast, only so much responsibility can be placed at the industry's door over the issue of e-waste, although it did itself few favours with some aggressive lobbying against proposed regulations on hazardous materials. To a public shocked by images of children in the Far East breaking up discarded TVs and PCs, the business made itself look dirty just as its consumer marketing was trying to convey the opposite impression.
However, one area where electronics is embracing calls for a more environmentally conscious approach is manufacturing. Its problem here though is that its own record for innovation can prove to be at odds with its environmental and economic objectives.
E&T reported earlier this year on the keynote by Dr Oh-Hyun Kwon, president of Samsung's semiconductor operations, at the International Solid State Circuits Conference (see http://bit.ly/OhHyunKwon). His main points were that chipmakers' fabs still use 40TWh of energy every year, more than the state of Michigan with a population of 10 million people, and that all the sector's players need to do something about it. The subsequent three months have underlined how timely Dr Kwon's comments were.
'The [power] consumption in 2008 was 47 per cent larger than that of 2001 and demonstrates a 7 per cent annual growth rate for the last two years,' he said. During the latter period, manufacturing processes were ramping up towards – and remain very much in the roll-out of – the latest process shrink, this time to the 28/32nm node.
Over the course of 2011, the industry is continuing to invest heavily in fab capacity, so much so that analysts at Gartner recently warned that it could be headed for an over-supply in spite of the disruption caused by the earthquake in Japan. SEMI, the trade association for the chip manufacturing chain, confirmed the spending spree in its most recent forecast.
'Total spending on fab projects could approach $47.2bn this year, above the estimated $38.6bn spent in 2010,' said Christian Gregor Dieseldorff, senior analyst in its Industry Research and Statistics group.
'Some companies will spend record amounts in 2011, reaching historic record levels. For example, [the world's largest foundry] TSMC increased capex from a record $5.9bn in 2010 to another record high of $7.8bn in 2011. Intel increased capex from $5.2bn in 2010 to $9.0bn in 2011. [The number two foundry] GlobalFoundries doubled its 2010 capex from $2.7bn to $5.4bn in 2011.'
According to Dr Kwon's model, more money and more fabs implies still more energy consumption. However, this is only part of the story. 'Most spending is directed towards upgrading existing facilities,' Dieseldorff continued.
A primary driver here is that fab owners want to avoid oversupply. The trend away from chip companies producing their own silicon to using third-party foundries has made those foundries much more sensitive to risks when their fabs are not operating at close to capacity. So incremental projects are, for now, more attractive than spending billions on greenfield factories.
However, energy costs are also playing a role in these capex plans. Whilst saving the planet is a desirable goal, saving money spent with utilities has become an equally compelling one.
In April, Intel again secured some green kudos with the announcement that it has received a Leadership in Energy and Environmental Design (LEED) 'silver' certification under the US Green Building Council-administered scheme for its Ocotillo operations in Chandler, Arizona.
The award was in respect of a long-established – rather than a brand new – facility, and is a demonstration of what the semiconductor industry can do already. As well as offices and support, Ocotillo houses three generations of fab.
'Given the size of the Ocotillo campus, it was an immense undertaking to seek certification of this manufacturing campus,' said Brian Krzanich, senior vice president and general manager of manufacturing and supply chain for Intel.
Specific steps taken to 'green' Ocotillo included the location of 200kW and 300kW solar arrays in a former car park; the build-out of infrastructure to recycle 90 per cent of solid waste (more than 9,000 tonnes) and achieve 66 per cent site-wide water conservation (saving 19 million litres of fresh water per day); and the installation of a system to capture storm water for use on site.
'The Ocotillo campus also utilises a pipe that feeds water not suitable for drinking from the City of Chandler's waste-water treatment plant directly back to Intel,' added Krzanich. 'As a result, 100 per cent of the irrigation water and 95 per cent of the cooling tower water is non-potable.'
Business case for green
Intel's LEED certification in Arizona is no one-off. In 2010, the company also received awards for sites in Haifa, Israel and Kulim, Malaysia. Nor is it alone in its enthusiasm for that programme. TSMC also has LEED awards – as well as recognition under Taiwan's Ecology, Energy Saving, Waste Reduction and Health scheme – for two of its newer factories: Fab 14/Phase 3 and Fab 12/Phase 4.
This work is largely being undertaken because it also saves money. 'We take these steps around the world because we see a combination of economic advantages and opportunities to reduce our environmental impact, which in turn betters our business,' said Krzanich.
According to metrics developed by the Semiconductor Industry Association, Ocotillo uses 26 per cent less energy than a standard chip manufacturing site.
'Where there is a clear return on investment for integrating individual 'green' features into our building design, it's easier to make it part of the design specifications for all building,' Ted Reichelt, Intel's principal environmental engineer and environmental strategies manager, has said.
Beyond Intel, TSMC and Samsung, one of the other clear signs that this LEED creed is being more widely adopted comes from International Sematech, the manufacturing think-tank that serves all fab owners.
In June 2009, it set up an Environmental Safety and Health (ESH) division within its ISMI manufacturing initiative. A relatively small-staffed project initially, it promotes technologies and strategies that will help companies reduce power consumption. Some of these involve large scale renewable installations and recycling, but it also backs transitions to LED lighting and more efficient heating, ventilation and air-conditioning.
Others seek to develop key performance indicators and ISMI-ESH launched TEECalcII last November, a software tool that tracks the total energy performance of fab equipment (the software is available for public download from www.teecalc.com).
ISMI-ESH estimates that such upgrades can cut utility bills by between 10 and 20 per cent and it is adding techniques to its existing standards arsenal that address, for example, the idle-mode consumption of vacuum pumps and point-of-use abatement systems used in chip manufacturing.
At ISSCC, Samsung's Dr Kwon noted that much fab plant still consumes 73 per cent of full operational energy in standby mode. Cutting this to just 50 per cent could save 2.3TWh of power across the whole sector every year. That saving alone could represent a 6 per cent cut in energy use.
However, his speech also began to look towards another impending challenge for chipmakers. Although companies have begun to take steps to control the existing infrastructure, overall power consumption continues to rise. This is where innovation becomes 'the enemy'.
It is now widely accepted that the 'classical' scaling of semiconductor dimensions that underpin the economic target in Moore's Law – doubling density and performance every 18 months or so – has come to an end. It simply comes down to the laws of physics – in the classical sense, there isn't that much more room at the bottom. At the same time, the power consumption of an end device has become as important as its performance.
Paying the consequences
All this has forced the industry to turn to new materials and innovations in areas such as lithography to meet goals for ongoing profitability. That in turn often means more sophisticated, more power-hungry production equipment. In addition, it is becoming ever more expensive to build the fabs themselves. GlobalFoundries has estimated that a state-of-the-art factory today costs a minimum of $4bn to build. The implications for ROI are obvious, and there is a shift on the horizon that is set to lift that price tag even higher.
Today's advanced fabs produce chips on 300mm wafers. Increasing wafer size while reducing device dimensions and device size equals still greater profitability but, at the outset, it is a big bet and after that the consequences are considerable. Toshiba, the world's third largest chipmaker, estimates that the last increase in wafer size recast the balance in terms of power consumption.
At fabs running 200mm wafers, 53 per cent of energy use was accounted for by the fab facility and 47 per cent by the manufacturing equipment within it. At 300mm, that changed to 46 per cent consumption for the facility and 54 per cent for the equipment. Now, silicon is getting ready to shift from 300mm to 450mm. The first fabs capable of taking the new size are going into construction today and actual production runs are due in two years' time, according to SEMI.
'SEMI's World Fab Forecast data identifies seven facilities (R&Ds, pilots and volume fabs) in the near future that are candidates for 450mm readiness,' said Dieseldorff. 'The first facilities are expected to come on line in 2013, though it remains to be seen if enough mature 450 mm tools will be available to fully equip a high-volume fab.'
This provides an important context for Dr Kwon putting energy consumption back on the agenda even though his industry is already working to manage everything that goes around the plant that makes the chips themselves.
'The semiconductor industry is reviewing the main causes of energy inefficiency in manufacturing equipment and processes. Controlling excessive operating margins in equipment and utility systems, along with the development of the next-generation process, can lead to a more energy-efficient manufacturing flow,' he said, but immediately noted that more work is necessary.
'We propose 'Interdisciplinary Collaboration' to capture all the possible energy saving opportunities in the areas relevant to the semiconductor industry. Interdisciplinary Collaboration in this context refers to the broad cooperation, not only within the semiconductor industry, but also with the other relevant industries including software, system, and infrastructure. We can further expand the scope of collaboration to academia and industry level consortiums such as WSC [the World Semiconductor Council] and ISMI. We believe all these eco-friendly Interdisciplinary Collaboration efforts will initiate and accelerate industry-wide energy saving activities, and can make a tremendous contribution towards the relief of the current environmental crisis.'
One caveat needs stating about the move to 450mm: only a handful of companies are likely to adopt the new size, given that it will best suit chips that are made in massive volumes and which are each likely to cost more than $100m to design. Nevertheless, the obvious candidates – Intel, Samsung (particularly as the world's largest manufacturer of memories) and the two leading foundry groups – will account for an increasingly large proportion of overall output.
However it is sliced, though, the broad collaboration envisaged by Dr Kwon will be essential. The challenge is balancing the extraordinary technological demands imposed by continuing on a Moore's Law-like path in a timely way with equipment that can deliver the innovations necessary on tighter power budgets. That conundrum is still some way from being resolved, although the willingness to do something about it is there.