Motors for semiconductor inspection mechanisms

Inspection mechanisms are specified in arcseconds. Encoder counts at the motor divide by the ratio to look impressive and tell you nothing about repeatability across a reversal.

Wedge
Industrial instrumentation and automation
Catalog reviewed
2026-08-18
Configurations screened
32 published

Filter and objective turrets, aperture wheels, polarizer rotations, notch aligners, focus drives and small stage axes. Each is a low-torque, low-speed positioner that has to arrive at a defined position, hold it while an image is taken, and repeat it thousands of times a shift. Some of them are inside a vacuum or a cleanroom, which constrains lubricant, materials and any wear mechanism that generates particles.

The mechanical load is small and easy: bearing drag plus a detent or index spring plus the inertia of a small wheel. The relation that governs the design is not a torque relation at all, it is a resolution and error budget. Output resolution is 360° divided by (counts per motor revolution × ratio), where counts per motor revolution is the encoder's pulses per revolution times four for quadrature decoding. Output repeatability, which is the number the specification actually contains, is the root-sum-square of geartrain lost motion, mounting compliance, encoder hysteresis and thermal drift, and lost motion normally dominates it by two orders of magnitude.

What normally controls the selection

Repeatability across a reversal, and whether the feedback can be placed where the tolerance lives. Everything else on an inspection mechanism has margin. The reduction ratio gets chosen to buy resolution and then buys lost motion at the same time, which is why increasing the ratio past the point where resolution is adequate makes the mechanism worse rather than better. The second governing factor is the feedback itself: the reviewed catalog holds one published encoder, a 12 pulse-per-revolution magnetic incremental with a reviewed fit to one motor platform. Any other combination is a configured part, and that has to be known before a specification is written around it.

Continuous, peak, starting and abnormal

Sizing on one of them is how a mechanism gets a part that works on the bench and fails in the field.

Row Continuous Sizes the thermal case Peak Sizes the torque case Starting Sizes the transient Abnormal Sizes the protection
What the mechanism is doing Effectively nothing. A turret index draws a fraction of the continuous rating for under a second, several times a minute.Detent breakout on an indexed turret, and the acceleration of the wheel inertia. Both are small relative to any planetary gearhead rating.The reversal. Every time the mechanism approaches from the other direction it traverses the full lost motion of the train before the load moves, and if the control loop closes on a motor-shaft encoder it will report arrival before the load has arrived.A turret driven into an obstruction, or a mechanism homed against a hard stop by design. Homing against a hard stop is common on these mechanisms and it means the drive stalls deliberately on every power-up.

The specifications that get compared are rarely the ones that decide the outcome.

What follows is what actually fails, and which published number sends people the wrong way.

Common failure modes
Failure mode Physical cause What it looks like in the field
Repeatability that passes unidirectionally and fails bidirectionallyGeartrain lost motion, with the encoder on the motor shaft. The control loop cannot see it.A fixed offset between clockwise and counterclockwise approaches, equal to the backlash referred to the output, stable over time and repeatable per unit.
Focus or image drift over the first hourThermal expansion of the mount and the mechanism as the instrument and the motor come to temperature.Drift that correlates with elapsed time from cold start, not with cycle count, and that disappears when the drive is left energized overnight.
Particle counts rise near the mechanismBrush and commutator wear, grease throw from the first stage, or a wearing detent. All three are point sources close to the sample.Particle excursions correlated with mechanism actuation rather than with airflow or with door events.
Homing position moves over lifeThe mechanism homes against a hard stop and stalls into it every cycle, and the stop or the output stage deforms.A slow, monotonic shift in the reference position, with the shift proportional to accumulated home cycles.
Specifications that mislead here
Specification Why it misleads in this application
Encoder resolution referred to the outputA 12 pulse-per-revolution encoder behind a 455:1 reduction gives 21 849 counts per output revolution, which is 59 arcseconds per count. That is a display number. The lost motion in the same train is typically 1 to 2 degrees, which is 61 to 121 times coarser.
Gearhead backlash quoted as a typical valueTypical is not a tolerance. On a mechanism specified in arcseconds the distribution matters more than the mean, and it changes with run-in and with load direction.
Reduction ratio as a proxy for precisionMore reduction improves resolution and degrades lost motion at the same time. Past the point where resolution is adequate, more ratio makes repeatability worse.
Motor class as a proxy for cleanlinessBrushless removes the brush wear source and leaves the grease, the bearings and the gear wear untouched. Vacuum and cleanroom compatibility is a lubricant and materials question that no motor class answers by itself.

The tradeoffs that matter here

Spur or planetary reduction

Row Spur Planetary
When it is right Lower ratios, where a shorter train means less accumulated lost motion and a simpler error budget.Higher ratios in a small diameter, and where the load sharing between planets averages some of the tooth-to-tooth error.
What it costs Lower torque rating, and single tooth-pair contact at each mesh, so individual tooth error is not averaged.Four stages of lost motion in series. The published four-stage ratios reach 455:1, and every stage adds to the reversal error.

Brushed coreless or brushless

Row Coreless brushed Brushless
When it is right Low duty, outside the cleanroom boundary, where brush particulate is contained and the total running hours are low.Inside the tool, where particulate is a specification, and on mechanisms that move continuously rather than indexing.
What it costs A particle source and an electrical noise source close to the sensor. Commutation ripple appears as velocity ripple during a scan.A driver whose switching is an EMI source next to a sensitive detector. The published brushless platform is 24.2 mm, which is large for a turret.

Where the feedback goes

Row Motor-shaft encoder Output or load-side encoder Hard-stop or index reference plus open-loop steps
When it is right Velocity control and stall detection, and where a unidirectional approach removes lost motion from the budget.The repeatability specification is tighter than the train's lost motion and a unidirectional approach is not acceptable.The mechanism has a small number of discrete positions, such as a filter turret, and each has a mechanical detent.
What it costs Blind to everything between itself and the load. The published encoder is 12 pulses per revolution with a reviewed fit to one platform.A configured part in every case. Mounting it stiffly to the load is the hard part, not the encoder.The detent, not the drive, sets the repeatability. The drive has to stall into it without damaging anything.

Worked example

Check whether a filter turret specified at 0.02 degrees can be closed with the published feedback.

Position tolerance
±0.02°, about 72 arcseconds
Turret
six positions, indexed
Published encoder
12 pulses per revolution, magnetic incremental, 2 channels
Encoder maximum frequency
20 kHz
Candidate ratio
455.19:1, four-stage planetary
Calculation
Step Expression Result
Counts needed for the tolerance360° ÷ 0.02°18 000 counts per output revolution
Ratio needed at 48 counts per motor revolution18 000 ÷ 48375:1
Counts at the candidate ratio48 × 455.1921 849 counts per output revolution
Resolution at the output360° ÷ 21 8490.0165°, or 59.3 arcseconds per count
Encoder frequency at the rated point23.5 rpm × 455.19 = 10 697 rpm; 10 697 ÷ 60 × 122 139 Hz per channel, 11 percent of the 20 kHz maximum
Encoder frequency at no load27.9 rpm × 455.19 = 12 700 rpm; 12 700 ÷ 60 × 122 540 Hz per channel, 13 percent of the maximum
Assumed lost motion expressed in counts1.5° ÷ 0.0165° per count91 counts
Lost motion against the tolerance1.5° ÷ 0.02°75 times the tolerance

Result

The resolution closes and the repeatability does not. At 455.19:1 the published encoder gives 59.3 arcseconds per count against a 72 arcsecond tolerance, and the encoder frequency is an eighth of its maximum, so the feedback is comfortable. The assumed 1.5 degrees of lost motion is 91 counts, 75 times the tolerance. Nothing about the encoder choice changes that. The mechanism closes only if it always approaches from one direction, or if the turret indexes into a detent that defines the position mechanically, or if the feedback moves to the output shaft. There is a second gate before any of that: the reviewed catalog holds one published encoder with a reviewed fit to one motor platform, and MM-C1626-P04552-240A is not that platform. This mechanism needs a configured feedback assembly and a measured lost-motion figure, so it goes to an application review rather than to a published part number.

What would change it

A lower ratio reduces lost motion and reduces resolution together. At 157.46:1 the resolution falls to 171 arcseconds per count, which fails the tolerance outright, so the ratio cannot simply be lowered. A higher-resolution encoder changes the resolution line and none of the others. Moving to a unidirectional approach removes lost motion from the budget entirely and is usually the cheapest fix, at the cost of an extra move on every reversal. A measured lost-motion figure is the single most valuable input here, because the whole conclusion turns on an assumed number.

Evidence, in the order it is worth collecting.

Each step names the measurement, not the intention.

  1. Measure lost motion on the assembly

    Applied torque against measured output angle, both directions, on at least ten units. Report the distribution, not a typical value.

  2. Measure bidirectional repeatability at the load

    An external angular reference at the turret, not encoder counts, over several hundred cycles from both directions.

  3. Characterize thermal drift

    Position at the load from cold start to thermal equilibrium, with the mechanism at its real duty. Separate motor self-heating from ambient.

  4. Establish the particulate contribution

    Particle counts at the mechanism with the drive cycling and with it stationary, at the tool's own cleanliness class.

  5. Qualify the homing method

    If the mechanism homes into a hard stop, cycle it to the full program life and measure reference drift and output-stage condition.

Send the positioning requirement for an application review.

Send the tolerance, the approach direction, the index rate and the cleanliness class. You get back a resolution and lost-motion budget, and a feedback configuration with the fit stated.

Send the positioning requirement for an application review