The Fundamental Architecture: How Each System Processes Information

Open-loop control executes a fixed, pre-programmed sequence of commands — a recipe — without ever measuring whether the intended outcome was actually achieved. The controller sends a signal to the actuator, the actuator acts on the process, and the sequence ends there. No sensor reads the output. No correction is made. The system’s accuracy depends entirely on how well the recipe was calibrated and how stable the process environment remains between calibrations.

Closed-loop control adds a feedback path. A sensor continuously measures the actual process output — temperature, film thickness, etch depth, chamber pressure, or another critical variable — and compares that measurement to the desired setpoint. The difference, called the error signal, is fed into a controller algorithm that calculates a corrective adjustment to the process input. This cycle repeats, typically hundreds or thousands of times per second in real-time implementations, driving the error toward zero.

The information flow in each architecture is structurally different. In an open-loop system, information flows in one direction only: from the recipe specification, through the controller, to the actuator, and into the process. In a closed-loop system, information flows in a continuous loop: from the process, through the sensor, back to the controller, which compares it against the setpoint and adjusts the actuator accordingly. This loop is what gives closed-loop control its ability to self-correct.

The absence of a feedback path in open-loop systems is not merely a limitation — it is also a design choice. Removing the sensor and feedback loop reduces system complexity, eliminates sensor-induced latency, and removes the risk of control instability. For processes that are inherently stable and well-characterised, open-loop control can be the more appropriate and cost-effective architecture.

Precision, Disturbance Rejection, and the Cost of Feedback

The precision achievable by each architecture is determined by the same factor: how much the actual process output deviates from the intended output. Open-loop control achieves precision only when the process is perfectly predictable — when equipment behaviour, ambient conditions, incoming material properties, and consumable state all remain constant between calibration and execution. In semiconductor manufacturing, this is rarely the case for extended periods.

Disturbance rejection is the key capability that distinguishes closed-loop from open-loop control in practice. A disturbance is any unplanned change that affects the process output — a gradual shift in RF power delivery as a match network ages, a slight variation in gas flow caused by upstream pressure fluctuations, or a change in wafer thermal conductivity due to incoming film stack differences. Open-loop control is blind to all of these. Closed-loop control detects their effect on the output and corrects for them automatically.

“Closed-loop control does not prevent disturbances from entering the process — it detects their effect on the output and corrects before the deviation becomes a defect.”

The cost of adding feedback is real and must be weighed carefully. Sensors capable of measuring critical process variables in real time — optical emission spectrometers, in-situ ellipsometers, laser interferometers, high-speed pyrometers — add capital cost, require calibration and maintenance, introduce potential failure modes, and can add latency to the control loop. If the sensor measurement lags the actual process state by a significant fraction of the process timescale, the feedback signal may arrive too late to prevent overshoot, potentially destabilising the loop rather than stabilising it.

Controller tuning — particularly for PID controllers — is the mechanism by which the trade-off between response speed and stability is managed in closed-loop systems. The proportional gain determines how aggressively the controller responds to an instantaneous error. The integral term corrects for sustained steady-state errors. The derivative term anticipates the rate of change and damps overshoot. Poorly tuned PID parameters in a semiconductor process tool can cause oscillation in chamber temperature, pressure, or gas flow that is more damaging to process uniformity than the disturbances the loop was designed to reject.

Where Each Architecture Is Used in Semiconductor Equipment

Semiconductor process equipment applies open-loop and closed-loop control selectively across different subsystems and process types, based on the precision requirements, sensor availability, and process timescales of each application. No modern process tool uses exclusively one architecture — the two are combined at the subsystem level.

Thermal Control Systems

Wafer temperature control is one of the most widely cited examples of closed-loop control in semiconductor equipment. In rapid thermal processing (RTP) systems and epitaxial reactors, pyrometers or thermocouples measure wafer or susceptor temperature and feed that signal to a PID controller that adjusts lamp power or heater current. The closed-loop architecture is essential because thermal mass, emissivity variation, and lamp aging all cause the relationship between heater power and wafer temperature to drift over time. According to IEEE publications on semiconductor manufacturing, temperature uniformity across 300mm wafers is a critical yield driver, making feedback control indispensable in thermal processes.

Simple furnace anneals with slow ramp rates and stable, well-characterised thermal loads can operate with open-loop control — the thermal mass is large enough that disturbances cause only slow, manageable drift, and the cost of adding high-speed pyrometry to every zone is not always justified for legacy process nodes.

Plasma Etch and CVD Processes

Plasma etch tools use a combination of open-loop and closed-loop control. Chamber pressure, gas flow rates, and RF power levels are typically controlled in closed-loop using pressure transducers, mass flow controllers, and RF power monitors. Etch endpoint detection — determining precisely when the etch has reached the target depth or cleared a film — is a closed-loop function implemented using optical emission spectroscopy (OES) or interferometry. The endpoint signal closes a loop that terminates the etch step, preventing over-etch.

Chemical vapour deposition (CVD) processes similarly rely on closed-loop mass flow control to maintain precise gas mixture ratios, while deposition rate and film thickness may be controlled in an open-loop fashion based on time and calibrated rate data, with in-line metrology providing run-to-run feedback rather than in-situ real-time correction. Standards bodies such as SEMI publish equipment interface standards that define how metrology data is communicated between tools and APC systems for exactly this purpose.

Lithography and Stage Positioning

Lithography scanners represent the most demanding application of closed-loop control in semiconductor manufacturing. The wafer stage positioning system uses laser interferometry or encoder feedback to achieve nanometre-level positioning accuracy at high throughput. The lens aberration correction system uses wavefront sensors and actuators to maintain focus and overlay performance. These are high-bandwidth closed-loop systems where the feedback loop operates at kilohertz frequencies to compensate for vibration, thermal expansion, and mechanical disturbances in real time. As documented by SPIE in its proceedings on optical lithography and microlithography, stage servo bandwidth and overlay correction are among the most technically demanding control problems in the industry.

CMP and Wet Process Equipment

Chemical mechanical planarisation (CMP) tools use closed-loop control of platen pressure, carrier head pressure, and slurry flow rate to maintain consistent material removal rates. In-situ endpoint detection using optical or eddy current sensors closes the loop on removal depth, stopping the polish when the target thickness is reached. Wet clean and etch tools typically use open-loop control for chemical dispense volumes and process times, relying on the stability of chemical concentration and temperature to maintain consistent outcomes.

Advanced Process Control: Closing the Loop at the Fab Level

Advanced process control (APC) extends the closed-loop concept from individual equipment subsystems to the entire fab production system. APC operates at three timescales — in-situ real-time control within a single wafer process, run-to-run (R2R) control between wafer lots, and lot-to-lot control across production campaigns — each addressing a different class of process variation.

The distinction between in-situ closed-loop control and run-to-run control is important. In-situ control corrects disturbances within the timescale of a single wafer process — milliseconds to minutes. Run-to-run control corrects slower drifts that are only observable across multiple wafer runs — hours to days. Both are forms of closed-loop control, but they operate on different feedback timescales and address different sources of variation.

Fault detection and classification (FDC) is a complementary APC capability that monitors equipment sensor streams in real time and detects anomalous patterns that indicate equipment degradation or process excursions. FDC does not directly adjust process parameters — it alerts engineers or triggers automated responses such as wafer holds or tool lockouts. The SEMI E10 and E116 standards define frameworks for equipment reliability and FDC data collection that are widely implemented across the industry.

Model predictive control (MPC) represents a more sophisticated closed-loop architecture increasingly used in semiconductor APC. Unlike PID control, which reacts to the current error, MPC uses a mathematical model of the process to predict future outputs and optimise the control input sequence over a receding time horizon. MPC is particularly suited to processes with significant time delays, strong interactions between multiple controlled variables, or hard constraints on actuator limits — characteristics common in multi-zone thermal systems and multi-step etch sequences.

Choosing the Right Architecture for Your Process

The selection between open-loop and closed-loop control for a given semiconductor process application is a structured engineering decision, not a default preference for one architecture. The relevant criteria are the process sensitivity to disturbances, the availability and reliability of real-time sensors, the process timescale relative to achievable sensor latency, and the cost-benefit ratio of adding feedback instrumentation.

Open-loop control is the appropriate choice when: the process is highly repeatable and well-characterised over the production lifetime of the equipment; the dominant disturbances are slow enough to be managed by periodic offline calibration; the sensor required for closed-loop operation is not available, not reliable in the process environment (e.g. high-temperature plasma), or adds cost disproportionate to the precision benefit; or when the feedback loop itself introduces instability risk that outweighs the disturbance rejection benefit.

Closed-loop control is required when: the process output must be maintained within tight tolerances despite equipment aging, chamber condition changes, or incoming material variation; the process involves a discrete endpoint that must be detected in real time (etch endpoint, CMP endpoint, deposition thickness target); the process timescale is short enough that offline calibration cannot keep pace with drift; or when the consequences of an out-of-spec wafer — in terms of yield loss, rework cost, or downstream process impact — are high enough to justify the sensor and control system investment.

In practice, most high-precision semiconductor process tools implement a layered control architecture: closed-loop control at the subsystem level for critical variables (temperature, pressure, gas flow, RF power), open-loop recipe execution for the overall process sequence, and run-to-run APC at the fab level to correct for inter-run drift. This layered approach captures the benefits of feedback where it matters most while preserving the simplicity and throughput advantages of open-loop execution where feedback is not warranted. Research published by IEEE and documented in proceedings from the Electrochemical Society consistently identifies this layered architecture as the standard approach for sub-20nm process nodes.

As semiconductor technology continues to scale, the boundary between open-loop and closed-loop control shifts progressively toward feedback. At each new technology node, the process windows narrow, the tolerances on critical dimensions tighten, and the number of variables that must be actively controlled to maintain yield increases. What was adequately managed by open-loop control at one node may require in-situ closed-loop feedback at the next. This trend drives continued innovation in sensor technology, control algorithms, and APC software — areas that PatSnap Eureka users can track in real time across the global patent landscape through PatSnap’s innovation intelligence platform.