Chip away at avoidable waste
November 1st 2004

Vilnis Vesma, senior consultant with National Industry Fuel Efficiency Service writes about effective energy targeting and monitoring in the microelectrics sector

The microelectronics industry features many energy-intensive process steps like electric furnaces and plasma etching (with their associated nitrogen, vacuum, and other energy-consuming auxiliary supplies), and yet energy demand is often only weakly related to production throughput because of a generally conservative approach to standing down idle production tools.

Instead, demand for energy in microelectronics fabrication is predominantly climate-related. The internal space has to be kept supplied with a steady flow of ultra-clean air at tightly-controlled temperature and humidity, with exhaust scrubbers to remove all traces of the hazardous chemicals used in the processes. Indeed, if you think of a fab as a massive air handling unit with a factory attached, you will not be far from the truth.

Energy demand that depends on the weather is notoriously difficult to manage because it is so variable. Hot weather necessitates chilling, cold weather calls for heating, and humidity correction often requires both gas and electricity. In order to detect energy waste (and hence cut consumption) the energy manager needs to take account of these influences, measuring them so as to calculate the target (i.e, expected) consumption in each stream on a weekly assessment cycle. He will for instance record a cooling ‘degreeday’ figure, this being a single number for the week whose value is proportional to kilowatt-hours of chiller demand. It is analogous to a production-volume figure or mileage in that it acts as the driving factor for energy consumption.

Each week, by comparing all his target consumptions with actual measured values, he can then calculate the discrepancy in each case, cost them out, and thereby present an overspend league table (OSLT) at the top of which he will find his most costly problems that week.

This routine scanning for new problems paid off handsomely in the case of one chip factory with its own liquid nitrogen plant.

Excess electricity consumption, relative to nitrogen output, showed up during 2003 and was traced to a second compressor being put into use to support the main unit.

The root cause, unknown at the time, was actually a problem with a non-return valve.

Having discovered what the operators had done (and why) the plant was able to rectify the fault which otherwise could well have persisted for weeks and wasted tens of thousands of pounds.

Prompt detection of significant hidden avoidable energy waste is assured by the use of the weekly OSLT. However, there is also the possibility that adverse changes in performance may have occurred before the targeting and monitoring regime is put into effect and such problems will often come to light during the setup phase of the targeting scheme. For example, excess gas consumption was found at one fab during target-setting. It was noticed that consumption relative to the weather had been significantly lower up to the previous March; the difference was a fixed, nonweather- related, continuous 360kW demand. Consulting the diary showed that two of the fab’s three heating boilers had been repaired during March and put back into service. The 360kW difference was the standing heat loss from the excess boiler plant. Clearly, with security of service being a key requirement, it was undesirable to rely on a single boiler; but at least they could limit themselves to one hot standby unit, rather than two, and thereby cut 180kW off the standing demand.

Whether the energy waste gives itself away during the setup phase, or after the targeting and monitoring regime is put into effect, the method relies on there being some observable change in behaviour. So people often ask what happens about plant and equipment that is, and always has been, operating wastefully. The answer is that if there are two or more comparable buildings or processes, these activitybased targeting methods can provide valuable insights through an advanced form of benchmarking. Take the case of eight air compressors operated by one company in this sector. Most of them displayed a very tight correlation between electricity used and compressed air delivered (figure 1 is typical) but one compressor had the anomalous characteristic in figure 2, which shows that it uses electricity at zero air output. This is symptomatic of leakage before the air meter.

Straight-line relationships like those shown in figures 1 and 2 are a surprisingly powerful way of viewing how a process or building uses energy, but they only work when there is a single driving factor such as the cooling degree day value. In semiconductor manufacture, it is not uncommon to have two or more influences at work.

Figure 3(a) shows the prediction error for a chiller system when assessed against cooling degree days only, while figure 3(b) shows the improvement when the influence of humidity is accounted for as well. How? Each week, the cooling degree value and the dehumidification demand index were calculated from real-time weather-station measurements, and the expected consumption for the week was calculated as follows:

2 000 000 kWh per week fixed overhead 1 000 kWh per cooling degree day 100 kWh per unit of dehumidification demand

Tabular targets of this form can be extended to any number of driving factors including production throughputs (typically reckoned in this industry by the number of wafer moves).They incidentally provide the foundation for an effective method of correcting for product mix and output something that many companies in climate change agreements may be interested in.

More articles from Nifes Consulting Group: