News

Machining to 2 Microns: What Tolerance That Tight Actually Requires

Aug 27, 2026

±0.00008 inches. Two microns. For scale: a human hair runs somewhere around 70 microns across.

Machining to 2 Microns: What Tolerance That Tight Actually Requires

±0.00008 inches.

Two microns. For scale: a human hair runs somewhere around 70 microns across. The tolerance band is a small fraction of the width of a single hair — and it has to hold across every part in the run, not just the one you measured.

Plenty of shops will quote it. Meaningfully fewer will hold it, document it, and still be in band on part four hundred.

The gap between quoting and holding is where this article lives. If you specify tight tolerances — or you're evaluating a supplier who claims them — here's what's actually required.

Tolerance Gets Expensive Non-Linearly

The first thing to understand is that the cost curve isn't a line.

Going from ±0.005″ to ±0.0005″ is a real change but a manageable one: better machines, more attention, some added inspection. Going from ±0.0005″ to ±0.00008″ is a different category of problem. Roughly speaking, each halving of the tolerance band doubles the process-control burden — more inspection points, slower cycles, higher scrap exposure, tighter environmental control, more skilled labor per part.

That's why a drawing with a blanket tight tolerance on every feature costs so much more than one where tight tolerance is applied only where function requires it. The two drawings look similar. They are not similar parts to make.

Thermal Stability Stops Being Background

At normal tolerances, ambient temperature is irrelevant. At two microns, it's a primary variable.

Steel has a coefficient of thermal expansion around 11 to 12 microns per meter per degree Celsius. Run that on a 100 mm feature: roughly 1.2 microns of dimensional change per degree. A two-degree swing in shop temperature moves the part further than the entire tolerance band.

This is why dimensional metrology is standardized at 20°C, and why holding micron tolerances requires:

Climate-controlled inspection — not just a clean room, a temperature-stable one
Soak time — parts coming off the machine need to equilibrate before measurement, and larger masses need longer
Coolant temperature control, since coolant is in direct contact with the part during cutting
Machine warm-up cycles, because spindle growth during the first hour of operation is real and measurable

A shop that measures parts straight off the machine in an uncontrolled room isn't holding two microns. It's reporting numbers that happen to look like two microns.

Metrology Is the Actual Bottleneck

You cannot hold what you cannot measure, and this constraint binds earlier than most people expect.

The conventional guideline is that your measurement system should resolve to roughly a tenth of the tolerance band. For a ±2 micron requirement, that puts you in the sub-half-micron range — and resolution is the easy part. Measurement uncertainty, which includes fixturing repeatability, operator variation, and environmental drift, has to stay small relative to the band as well.

Gage R&R matters here in a way it doesn't at looser tolerances. If your measurement system consumes 30% of the tolerance band in variation, you're no longer measuring the part. You're measuring the part plus the noise, and you'll scrap good parts while passing bad ones.

This is also why we run CMM inspection at multiple stages of the process rather than only at final. At micron scale, discovering drift at final inspection means discovering it after the value is already in the part. Catching it in-process is the difference between an adjustment and a scrap report.

Work-holding Deforms the Part

At micron scale, clamping is not neutral.

Hold a part firmly enough to machine it and you've elastically deformed it. Machine it to size in that deformed state, release the clamp, and it relaxes into a shape that isn't the one you cut. On thin-walled or asymmetric geometry, that relaxation alone can exceed a two-micron band.

Which makes fixture design part of the tolerance stack rather than a separate concern. Low-stress workholding, distributed clamping force, and — critically — measuring the part in the free state rather than while it's still fixtured. A part that measures perfectly in the vise and fails on the surface plate is a workholding problem, not a machining problem, and diagnosing it requires knowing to look.

Tool Wear Inside a Single Run

Tool wear that's completely irrelevant at ±0.005″ becomes a dominant error source at ±0.00008″.

A carbide end mill doesn't fail suddenly; it degrades progressively. Over a run of several hundred parts, that progression will walk your dimension out of band long before anything sounds or looks wrong. The part coming off the machine at number 300 is not dimensionally identical to number 1 unless someone is actively compensating.

That means monitoring wear per-part rather than per-run, applying offset compensation on a schedule derived from measured drift rather than a tooling catalog, and — on the tightest work — accepting shorter tool life than the economics would otherwise suggest. Running a tool to the end of its useful life is a cost decision that tight tolerance takes away from you.

The Operator

There's a category of judgment that isn't in the program.

An experienced machinist hears a change in the cut before the metrology reports it. They notice a chip color that's slightly off, a coolant film breaking differently, a finish marginally duller on this part than the last. None of that is in the CAM file, and none of it shows up in inspection data until it's already a dimensional problem.

At normal tolerances this is a nice-to-have. At two microns it's load-bearing. The process holds because someone is paying attention to it — and that's not a claim any equipment list can make for you.

When You Actually Need It

An honest note, and one worth more to you than another capability claim: don't over-specify.

Micron-level tolerance is genuinely justified on sealing surfaces, bearing fits, optical mounts, hermetic seal interfaces, and mating features where accumulated stack-up would otherwise put the assembly out of function. In those places it isn't optional, and a supplier who can't hold it will fail you.

Everywhere else, it adds cost and lead time without adding function. A drawing that applies tight tolerance selectively — where function requires it, and nowhere else — gets you a better part, faster, for less.

The most valuable conversation to have with a machining supplier early in a design is which of your callouts actually need to be tight. A supplier who only ever tells you they can hold everything isn't giving you useful information.

We hold to two microns and micro-machine features down to .004″. We'd still rather help you specify the tolerance you need than quote the one you wrote out of caution.

BoldX Industries

NADCAP-accredited under AC7108. AS9100D certified. ITAR-registered. Precision machining, value-added assembly, and QPL-qualified circular hermetic connectors for MIL-DTL-5015, 38999, 83723, and 26482. Batavia, OH. U.S. owned and operated.

getboldx.com

If you are qualifying a connector supplier for a flight-critical program and want a walkthrough of our QPL scope, our most recent audit posture, and how we support first-article packages, you can reach us at getboldx.com/contact