Semiconductor Plant Energy Management: Meter First

One 30-minute interval can set the month's demand charge

2:05 p.m.: a chiller restarts while several production tools and air compressors are already running. The overlap lasts 30 minutes. That short event may set the month's maximum demand. Good semiconductor plant energy management starts by seeing this event at the incoming supply, then tracing it to the systems that caused it.

Electrical and electronics (E&E) products are Malaysia's largest export product group. MATRADE reported RM601.18 billion of E&E exports in 2024, a record, and 39.9% of everything Malaysia exported.

Investment is also adding capacity in the northern manufacturing corridor. Infineon opened the first phase of its Kulim 3 silicon carbide fab in August 2024, with a second phase of up to five billion euros planned.

Yet many plants still see only the Tenaga Nasional Berhad (TNB) incoming meter and the monthly bill. The gap between that total and equipment-level operation is where avoidable idle load, poor scheduling and utility faults can hide.

Semiconductor plant energy management must match the plant type

Start with the process. A wafer fabrication plant, an outsourced semiconductor assembly and test (OSAT) plant, and a general electronics factory do not have the same utility profile.

This table is a starting point, not a percentage split. Tool set, product mix, cleanroom class and operating hours can change the order. Do not copy an energy breakdown from another site.

A fab may need deionised (DI) or ultrapure water (UPW) metering near the top of its plan. A general electronics plant may get more value from line, oven and compressor meters. An OSAT site often sits between these profiles.

Meter utilities in a top-down hierarchy

Build the meter tree from source to use. Each lower level should explain a useful part of the level above it.

First, make the TNB incomer and main switchboard the reference. Record active power, energy and power factor. Meter the major distribution boards or transformers beneath it. Check that the sum of submeters is reasonably close to the incoming total after known losses and unmetered loads. A large unexplained balance means the hierarchy is incomplete or a meter needs checking.

Second, instrument the chilled-water plant as a system. Process cooling is often a major site load and deserves early meter coverage. Chillers alone do not show pump and cooling-tower cost. Pair electrical meters with flow and temperature sensors. This lets the team compare power with delivered cooling. For water without glycol, an approximate thermal load is:

$$ \text{Cooling load (kW)} \approx 4.186 \times \text{flow (L/s)} \times \Delta T\ (^\circ\text{C}) $$

A 100 L/s flow with a 5°C temperature difference delivers about \(4.186 \times 100 \times 5 = 2{,}093\) kW of cooling. Confirm sensor accuracy and fluid properties before using this result for formal measurement and verification.

Third, separate cleanroom HVAC from general comfort cooling. Track supply and return conditions, fan power, cooling, reheat, humidity and schedule. Air-change and pressure settings affect contamination control. Any change needs process, quality and safety approval. Cleanroom optimisation deserves its own engineering study.

Fourth, meter each major compressed-air, vacuum, exhaust and abatement train. Header pressure alone cannot show electrical cost. Compare input power with flow and production state. Our guide to compressed-air energy efficiency in Malaysia covers leak checks, pressure and controls in more detail.

Fifth, map water from plant inlet to treatment and high-use areas. In DI or UPW systems, measure feed, product, reject and make-up streams. Balance the flows. A change in recovery or an unexpected night flow can then be investigated.

Sixth, connect every utility trend to a production state. Without that link, a falling energy line may mean better control, lower output or a stopped line.

Treat maximum demand as an operating event

Under Regulatory Period 4 (RP4), effective 1 July 2025, TNB's medium-voltage demand charges are RM89.27 per kW for General supply and RM97.06 per kW for Time of Use (ToU) supply. The charge uses the single highest 30-minute demand interval in the month. TNB defines the ToU peak period as 2 p.m. to 10 p.m. on weekdays. Check the current schedule on TNB's tariff page before making an investment decision.

Consider a ToU plant whose billing maximum includes 180 kW of avoidable coincident load. Perhaps a standby chiller started while compressors were loaded and non-critical charging was active.

$$ \text{Demand-charge difference}=180\ \text{kW}\times\text{RM }97.06/\text{kW}=\text{RM }17{,}470.80 $$

The exposure for that month is RM17,470.80. This is not an energy-charge calculation and not a guaranteed saving. The plant saves this demand amount only if the action reduces the actual billing maximum. Moving a load to another already-high interval may simply create a new peak.

Use an alert below the site's agreed demand limit. When it triggers, operators need a prepared response list. Delay only approved loads. Adjust chiller staging within process limits. Avoid starting large systems together. The decision must protect output, product quality and safety.

See the fuller guide on how to cut TNB maximum-demand charges.

Correlate energy with idle, running and ramp states

A useful dashboard explains why consumption changed. Start with production tags that operators already understand: idle, running, ramp, changeover, maintenance and shutdown.

Then calculate a suitable intensity measure:

$$ \text{Specific energy}=\frac{\text{kWh in the period}}{\text{good units, panels or lots in the period}} $$

If a line uses 3,600 kWh to make 1,200 good units, its specific energy is \(3.0\) kWh per good unit. Keep idle energy separate. Dividing 500 kWh by zero output during an idle shift gives no useful result. Report the idle baseline in kW and total idle kWh instead.

Compare like with like. Product mix, yield, ambient conditions and planned maintenance can move the result. A higher kWh per lot is a question to investigate, not proof of poor performance.

This same logic supports a wider factory energy-monitoring plan. It also creates a defensible baseline for projects such as chiller sequencing, pressure reduction or schedule changes.

CobiNeural connects meters, utilities and operations

CobiNeural can collect equipment-level energy and water data, monitor maximum demand, issue alerts and show dashboards alongside production states. It can be deployed as a standalone platform or as an overlay on an existing building management system (BMS), programmable logic controller (PLC) or supervisory control and data acquisition (SCADA) system.

The practical design depends on the site. Existing signals can be reused where they are reliable. New submeters and sensors can fill important gaps. The platform also covers indoor air quality (IAQ), chilled water, billing, measurement and verification (M&V), Energy Efficiency and Conservation Act 2024 (EECA) reporting workflows, and International Organization for Standardization (ISO) 50001 energy-management reporting.

Software does not decide the hierarchy by itself. Engineers still need correct meter locations, current transformer ratios, naming, time alignment and data-quality checks. Start with the few meters that can explain the largest loads. Add depth only when it leads to an operating decision.

If you want to map the first metering points for your Malaysian plant, request a practical CobiNeural walkthrough.