Why Can't Chip Patterns Be Milled?

Turning, milling, and drilling share one feature: the tool has a pre-made edge — material is torn off the workpiece as chips by that edge. This differs from grinding, polishing, chemical-mechanical planarization, and ion-beam figuring: none of those has a “shape-is-the-answer” tool; material is removed layer by layer by abrasives, chemistry, or ions.

This article answers one question: why can’t chip patterns be milled? The answer has two layers. First, the dimensions a shop-floor miller or lathe can reliably deliver are bounded by three floors: tool-nose geometry, edge radius, and temperature. Second, the extreme end of the same family — single-point diamond turning — pushes shape accuracy below 100 nm and Ra below 1 nm, yet still fails three gates: the edge leaves the minimum chip at the nanometer scale, iron-based metals eat the diamond, and further figuring requires a different process family.

1. Four terms that must not be mixed

One cutter can produce a smooth surface while failing to make a 1 μm-wide groove — shininess is roughness, groove width is feature size.

A round-nose turning tool leaves an arc on the surface with every feed per revolution. When feed f is much smaller than nose radius R, the peak-to-valley geometric height is about f²/(8R); approximated as an arithmetic-mean profile, the common estimate is Ra ≈ f²/(32R). With R = 0.4 mm and f = 0.1 mm/rev, Ra ≈ 0.78 μm — the order of magnitude seen in general finishing turns. Drop the feed to 0.01 mm/rev with a 0.5 mm nose, and the same formula gives Ra ≈ 6.3 nm. The formula allows nanometers, but it computes tool-nose geometry only: vibration, edge plowing, lateral plastic flow, and grain steps are not in this term^[1].

Figure 1 | The two floors: nose ripple and edge radius. Click for full size.

2. The edge is a fillet, not a geometric line

Zoomed in, the so-called sharp edge is a fillet, with radius called the edge radius r. If the uncut chip thickness t falls below some t_min, the material is mostly plowed into the surface instead of becoming a chip; the plowed material springs back, and dimensions and roughness stop following the feed formula.

Ikawa et al. tied this minimum chip thickness to the edge radius in diamond turning: with a well-behaved cutting system, the minimum thickness is on the order of 1/10 of the edge radius, and experiments produced chips of about 1 nm^[3]. The floor is set by how sharp the edge is — not by the operator turning the feed one step smaller.

How sharp can a single-crystal diamond tool be ground in practice? Zong et al.’s grinding results: edge radius about 30–50 nm (depending on the crystal-face combination: one set 35–50 nm, another 30–40 nm); theoretically certain orientations can be re-edged to about 1–6 nm, and earlier practical assumptions had it at 70–80 nm^[4]. With a 40 nm edge and the 1/10 order, the minimum chip is a few nanometers — “cutting” thinner than that is mostly plowing.

The micro-milling review places conventional end mills above 1 mm in diameter and micro end mills between 1 μm and 1 mm, with micro-mill edge radii usually under 5 μm, reaching 20 μm^[5]. Once the edge is the same order as the chip thickness, the plowing that macro cutting can ignore becomes the dominant effect: a tool can be bought at 0.1 mm diameter without every cut being chip formation instead of material squeezing.

3. Machine and material: heat and iron-based metals

However sharp the tool, the machine itself is growing during the cut. Linear thermal expansion ΔL = αLΔT: engineering handbooks give steel about (10.8–12.5)×10⁻⁶ /°C and aluminum about 23×10⁻⁶ /°C^[7]. Taking 12×10⁻⁶ /°C for steel: a 300 mm steel scale rising 1 °C elongates 3.6 μm; a 1 m steel ruler is 12 μm; 100 mm of aluminum is 2.3 μm. “Machine resolution 0.1 μm” and “the part stays 0.1 μm under room-temperature fluctuation” are two different things — coolant, spindle, air conditioning, and the temperature the part brings from the previous operation all enter that 1 °C.

Material adds a chemical limit: diamond tools wear fast on iron-based materials and generally cannot turn steel, iron, titanium, or pure nickel — iron-based materials catalyze diamond graphitization even at room temperature^[1]. Optical aluminum mirrors, copper, electroless nickel, germanium, and some plastics can go through single-point diamond turning; mold steel cannot. This is a material pairing, not a machine class.

Lower feed does not guarantee better function: Li et al. turning Al6061, at feeds low enough that chips stopped being continuous, saw roughness fall from about 7 nm to about 5 nm while normalized reflectance dropped from about 90% to about 82%^[8] — a smoother surface, optically worse.

4. Where single-point diamond turning stops, on one scale

Single-point diamond turning (ultra-precision lathe + nanometer-edge single-crystal diamond) delivers, per the review, shape accuracy better than 100 nm and surface Ra better than 1 nm, for ductile materials (non-ferrous metals, plastics, some IR crystals); hard/brittle materials go through ultra-precision grinding: shape about 100 nm, Ra about 10 nm (fused silica, silicon carbide, ceramics)^[1].

Figure 2 | The scale ladder from a 0.1 mm end mill to silicon atom spacing. Click for full size.

Silicon’s lattice constant, the CODATA 2022 recommended value, is a = 5.431020511×10⁻¹⁰ m; the diamond-structure nearest-neighbor distance (√3/4)a = 0.235 nm^[6]. Using that spacing as the ruler:

A 0.1 mm tool is still about 425,000 times wider than a silicon atom spacing; a 40 nm edge spans about 170 nearest neighbors — a whole sheet of atoms stands on the edge, not one.

Optical surfaces stretch the gap further: eyeglasses about 2000 nm rms shape, 10 nm rms roughness; space optics about 20 nm and 0.5 nm; a deep-UV projection lens about 2 nm rms shape; an EUV reflector mirror must reach 0.1 nm single-side shape and 0.05 nm rms mid-spatial-frequency roughness — the EUV wavelength is 13.5 nm^[1]. The last step on these surfaces is not a lathe: grinding or diamond turning takes the front stage, then magnetorheological, air-bag, and computer-controlled polishing do local figuring, and elastic-emission, float polishing, or ion beams push roughness to the atomic layer. Silicon-wafer chemical-mechanical polishing reaches Ra of about 0.1 nm with 12-inch flatness about 1 μm; the 12-inch bare silicon wafers used for EUV must hold total thickness variation under 200 nm and mid-spatial-frequency roughness under 0.1 nm^[1].

Back to the opening question: chip patterns can’t be milled because lithography defines lateral patterns (where the photoresist stays, where it goes), and the line width follows wavelength, numerical aperture, and process factor — the light source walks from 193 nm to 13.5 nm, and transistor feature sizes go below 10 nm; mechanical cutting’s lateral features follow the tool diameter, production micro end mills stop at a few tenths of a millimeter down to 0.1 mm, and fine tools snap under lateral force before they ever meet an atom^[1]. Gates, fins, and metal line pitches are not the job of “a particularly small end mill”; machining’s role left in the fab is chambers, fixtures, heat sinks, and some package metal parts. Scope: the numbers here come from published reviews and tests, for the process family with a geometric edge. Unresolved limits: tools will keep getting sharper and machines more stable, but the boundary drawn together by edge, temperature, and atom spacing does not move just because the edge is sharper.