Saving Energy In The Fab

It’s not an exaggeration to say that integrated circuits are a critical component of any effort to reduce global carbon dioxide emissions. The most efficient engines depend on microcontrollers to optimize fuel consumption. Global shipping uses sophisticated simulators for load balancing and route planning. Computing power that once needed room-sized cooling units now fits in a battery-powered pocket device.

The American Council for an Energy Efficient Economy views energy efficiency as the third-largest energy resource available to the U.S economy, delivering the equivalent of 313 large power plants since 1990. And yet, an old adage observes that a billion of anything adds up to a significant impact. While electronic devices are more efficient than they’ve ever been, there are more of them every year. IC manufacturing’s share of global CO2 emissions is increasing as the industry grows. How can semiconductor fabs use the technology they so eagerly supply to the world to reduce their own impact?

Energy and emissions
Many discussions of sustainability treat “energy” and “emissions” as interchangeable. Technically that’s not correct. Energy from renewable sources produces much less CO2 than energy from fossil fuels. Even among fossil fuels, different sources have different environmental loads. On the other hand, fabs emit greenhouse gases other than CO2, some of them with a much longer atmospheric persistence. This article focuses specifically on fab energy use; emissions from other sources will be discussed later in this series.

The GHG Protocol, developed by the World Resources Institute (WRI) and the World Business Council on Sustainable Development (WBCSD), partitions emissions into three categories. Scope 3 includes emissions produced as a consequence of a facility’s existence, but outside the facility’s control. For a semiconductor fab, these would include emissions attributable to employee commuting, to product shipping, and to production and shipment of raw materials used by the facility. While these emissions are important when considering the impact of the industry as a whole, they are not considered here. Rather, this article focuses on energy consumption by the fab itself, regardless of the source of the energy. Purchased electricity or steam will produce Scope 2 emissions, while natural gas or other fuel burned onsite for heat produces Scope 1 emissions. (Waste gases produced by the process itself or by abatement systems also fall under Scope 1, but will be addressed later in this series.)

Where does fab energy go?
Energy per mask layer is a useful metric for comparing energy consumption across fabs and across products. Because the number of mask layers tends to increase with time, simply holding constant the energy per mask layer is not enough to maintain, much less reduce, the fab’s energy consumption. In fact, the industry’s shift from batch to single wafer processing has tended to increase energy consumption. Dry etch and deposition processes are among the most energy intensive in the fab.

As Andreas Neuber, director of environmental services at Applied Materials, and his colleagues explained, fab energy use includes energy used to power the process equipment itself. That can include heating a chamber or driving a laser, as well as energy consumed by utilities supplying the process equipment, such as by vacuum pumps and point-of-use abatement systems. It also includes energy consumed by fab infrastructure, from the ventilation systems to the UPW plant. These account for, respectively, 44%, 40%, and 17% of overall fab energy consumption.

The first component, energy consumed by the process itself, is largely defined by the thermodynamics of the particular process step. There’s not much fabs can do to change the activation energy of a dopant or the decomposition energy of a molecule. In some cases, though, fabs can consider energy consumption when deciding what process or chemistry to use in the first place. Often, such decisions require fabs to balance energy use against both process performance and other environmental impacts. For example, more aggressive cleaning chemistries might reduce the necessary process temperature but produce more toxic waste products. As a result, some of the process energy savings might be consumed by waste mitigation. Conversely, if a chemical is eliminated because of its environmental impact, the replacement might require a higher temperature process or additional process steps. In most situations, though, fabs are constrained by the requirements of the process. Potentially more important opportunities for energy efficiency lie elsewhere.

For example, MIT student Matthew Branham found that most process support components run all of the time, whether the tool is processing wafers or not. Pumps, purge gas, water bath heaters, all work to maintain a consistent process environment regardless of actual manufacturing status. Thermal abatement systems tend to run at the highest temperature that might be required. Even some chemistries can be processed with less heat.

Inefficient though they are, these practices were established for good reasons. A chamber might be idle now, but the next load of wafers might be imminent. Increases in cycle time or losses in yield caused by fluctuating tool environments can consume just as much energy as idled tools might save. Devices like the Applied Materials iSystem Controller allow users to throttle back pumps and purge gases during cleaning and similar operations. Even at high utilization, the controller can reduce abatement and dry pump operating costs by up to 10%.

Still, Shaun Crawford, sub-fab product line manager in the Applied Materials Applied Services group, agreed that most users are not eager to tamper with the baseline process environment. SEMI’s E167 standard supports the development of a true “idle mode” by specifying how the fab control system should communicate expected idle time to the process equipment. Equipment designers also are considering ways to share some pumps, heat exchangers, and other support components between process chambers, optimizing both energy consumption and fab footprint.

Infrastructure, the third contributor to the fab’s total energy consumption, is the one most amenable to energy saving technologies used in other industries. Lights can be converted from incandescent and fluorescent bulbs to LEDs with smart control systems. Waste heat from chillers can be used to heat water or the fab itself. Air conditioning units can be replaced with more efficient models. The specific steps that are possible or economical may depend on the facility. At many facilities, “easy” steps in these categories have already been taken. They provide immediate energy cost savings and have little or no impact on the process, so they are low hanging fruit for fabs trying to minimize energy use.

Energy conservation on the leading edge
As with many facility-level optimizations, a leading edge fab being designed from the ground up will have more flexibility than a legacy fab being updated to a different process or product type. Early plans for 450mm fabs paid significant attention to power use and sustainability. The F450C utility focus group found that a 10% reduction in process equipment power consumption could shave about 2% from facility construction capital costs, achieved through smaller chillers and other elements. The delayed introduction of 450mm wafers does not mean the industry can put energy efficiency on hold, but forces manufacturers to find power optimizations within their existing processes.

Intel has been one of the leaders in this effort. At the 2008 Electronics Goes Green conference, Tim Higgs and Todd Brady reported that the company had reduced energy use per unit of production by 4% a year since 2002, with a 5% per year reduction goal going forward. According to Brady, Intel’s director of global sustainability efforts, the company has invested $145 million since 2008, saving about 3.2 billion kWh of energy and about $340 million in energy costs. The specific changes run the gamut from LED lighting to heat recovery from chillers. Intel’s Chandler, AZ plant, Brady said, saves about $500,000 per year by using real time information from chillers to maximize efficiency.

Similar efforts at UMC have achieved a 5.42% reduction in total power use since 2012, with the largest share attributable to energy conservation by the production equipment. UMC’s electricity usage index fell from 14.2 MWh per square meter of silicon in 2013 to 13.32 MWh per square meter in 2015. Both companies are investing in renewable energy to reduce the emissions associated with their power use.

LEED certification of both fabs and office buildings is now common in the industry. Texas Instruments’ experience appears to be typical: most of the costs associated with LEED certification were due to efficiency improvements that would have been considered regardless. As energy costs continue to rise and processes complexity continues to increase, fabs are learning that improvements that benefit the environment help the bottom line, too.

Read the first part of this series: Making Manufacturing Sustainable For Chips

Related Stories
Running Out Of Energy?
A new report documents what happens if computing continues at its current rate of consumption.
Making Manufacturing Sustainable For Chips
First in a series: What is sustainability and why does it matter?
A Semiconductor Approach To Desalination
A new growth market for a very good cause.
Think Globally, Act Globally
Fabs consume huge amount of electricity – does the industry have a responsibility to act sustainably?