Semiconductor CNC Machining: How It Really Works

• 1
  Key takeaways
• 2
  Why Semiconductor CNC Machining Is Different from General Milling
• 3
  How Tolerance, Setup and Thermal Drift Interact
• 4
  Material Behavior in Vacuum and Clean Environments
• 5
  Surface Finish, Burrs and Particle Control
• 6
  Where Milling Reaches Its Limits
• 7
  How to Prove the Part Is Right Before It Ships
• 8
  Which Machining Route Fits Which Semiconductor Part

A semiconductor tool is not a normal machine frame. It sits inside a vacuum or a clean environment, it moves wafers worth more than the machine, and it must not add particles to the process. That changes what the drawing has to say. A bracket for an automotive line and a bracket for a load-lock door can look almost identical on paper. They are not the same part.

The first difference is the specification. Semiconductor CNC machining usually carries a flatness or parallelism callout instead of a simple size tolerance. A wafer stage that is 0.02 mm out of parallel will not hold focus across a 300 mm wafer. Size alone tells you nothing about whether the part will work.

The second difference is cleanliness. A blind hole full of chips, a burr on a gas line, or a tapped hole with trapped swarf will all release particles once the tool is pumped down. Deburring and cleaning are not a final step you bolt on. They have to be planned at the quoting stage.

The third difference is traceability. Equipment builders often need material certificates, inspection reports and a paper trail that follows the part from raw stock to shipment. If the shop cannot produce that, the cheapest quote is expensive.

• 1
  Geometry firstFlatness, parallelism and concentricity drive the process choice, not overall size.
• 2
  Cleanliness is a featureEdge breaks, blind-hole depth and surface finish decide particle shedding.
• 3
  Paper mattersMaterial certs and inspection reports are part of the deliverable.

A tolerance of ±0.005 mm is roughly one twentieth of a human hair. It is achievable, but only in a stable loop. The machine, the fixture, the tool and the workpiece all move as the cut heats them. A 100 mm aluminum part grows about 0.0023 mm for every 1 °C it warms, so a 5 °C rise eats half the tolerance before the tool touches the part.

This is why we check the part, not the machine. A machine that repeats to 2 μm on a test artifact can still produce a 15 μm part if the fixture flexes or the stock moves. Probing the datum in the spindle, then cutting from that datum, removes the biggest source of error.

Setup count is the other lever. Every re-fixture adds a new datum and a new stack of errors. A part with five faces to machine on a three-axis mill needs four or five setups. The same part on a simultaneous five-axis center needs one, plus a flip if the back face is complex.

Temperature control is not glamorous, but it is cheap. Letting a casting normalize overnight, roughing it, then finishing after a cool-down is often the difference between a part that passes and a part that is remade.

• 1
  Cut from a probed datumNever trust the vise jaw as your reference on a tight part.
• 2
  Fewer setups, less errorEach re-fixture adds roughly 5–15 μm of position error.
• 3
  Rough, rest, finishStress relief between passes protects thin walls and large plates.

Aluminum 6061-T6 is the default for chamber bodies, brackets and manifolds because it machines fast and anodizes cleanly. Hardcoat anodizing gives a wear surface that resists galling on sliding parts. Conductive anodizing keeps the part grounded where charge buildup is a risk.

Stainless 304 and 316L are chosen where corrosion resistance or low outgassing matters. They machine slower, work-harden easily and tend to burr on fine edges. 17-4PH (SUS630) is used for shafts and pins that need strength after aging. It moves during heat treatment, so the drawing has to say whether the finish cut comes before or after.

Copper and beryllium copper appear in RF paths and thermal straps. They are gummy, they grab the tool and they are hard to hold to tight flatness. Sharp, polished carbide and light depths of cut help. Beryllium copper dust is a health hazard, so the shop needs control measures, not just a good program.

Plastics like PEEK and POM show up in insulators and wafer-contact parts. They move with humidity and heat, and they cannot be cleaned the same way as metal. A PEEK part that is dimensionally perfect but has a rough saw-cut edge will still shed.

Particle generation is mostly an edge problem. A sharp 90° corner on a machined surface will chip and shed long before the flat face does. Specifying a 0.3 mm edge break on every external corner costs almost nothing and removes a whole class of defects.

Surface finish matters where the surface touches the process. Sealing faces for O-rings are typically held at Ra 0.8–1.6 μm. A finer finish is not automatically better: a mirror finish can trap no seal, while a rough one leaks. Internal gas paths are often held at Ra 0.2–0.8 μm to reduce adsorption.

Blind holes are a trap. A hole that bottoms out flat will hold a chip no matter how well you rinse it. Adding a small drill point at the bottom, or opening a cross-hole to the main volume, lets cleaning fluid reach the swarf.

Cleaning has to match the material. Aluminum can take an alkaline wash, some copper alloys cannot. PEEK and similar plastics tolerate limited solvents. We agree the cleaning method with the customer before the first chip is cut, because it sometimes changes the drawing.

• 1
  Edge break everythingA 0.3 mm chamfer on external corners cuts particle shedding.
• 2
  Match finish to functionSeal faces at Ra 0.8–1.6 μm, gas paths at Ra 0.2–0.8 μm.
• 3
  Never flat-bottom a blind holeLeave a drill point or a cross-hole so cleaning fluid reaches the bottom.

The maximum processing size on our largest machines is 4,000 mm, with travels of 4,000 × 400 × 150 mm on the long-bed platform. A part longer than that cannot be cut in one piece. It has to be split, bolted or welded, and each of those joins brings its own error and its own cleaning problem.

Aspect ratio is the other wall. A pocket 200 mm deep and 20 mm wide needs a long, thin tool. It will deflect, it will chatter, and the finish will suffer. Deep pockets are sometimes better roughed by milling, then finished by EDM or by a ground insert.

Thin walls below about 0.5 mm are hard to hold on a mill. The cutting force pushes the wall away, and the spring-back leaves a taper. Filling the cavity with a support medium or roughing with a small radial engagement helps, but sometimes the answer is a different process.

Finally, hardness. Parts above roughly 45 HRC are usually ground or EDM-cut after heat treatment, not milled. Trying to mill a hardened vacuum valve seat is a good way to scrap it.

• 1
  4,000 mm is the ceilingLonger parts need a joint, and joints need a plan.
• 2
  Deep narrow pockets chatterPast roughly 10:1 depth-to-width, switch to EDM for the finish.
• 3
  Above 45 HRC, grindHardened seats and valve bodies are a grinding job.

Inspection starts with the raw material. We check the certificate against the drawing callout before the first cut, because a wrong alloy found after machining is a scrapped part. In-process checks catch drift while there is still material to correct.

Final inspection is 100% before shipment on this class of part. CMM reports, roundness traces and surface finish readings are available on request. The point is not to produce paperwork for its own sake. It is to give the equipment builder a record it can hand to its own customer.

For prototype builds, we quote and return a free DFM analysis within 12 hours, and production can start within 24 hours. Prototype parts typically ship in 3–5 days. That speed only helps if the drawing is already sound, which is why the DFM step matters more than the machining step on the first build.

The historical late-delivery probability on our production orders is below 2%. We publish that number because schedule slips are the most common complaint buyers have about machining suppliers, and it is the thing we measure hardest.