How the Lenses of ASML Lithography Machines Are Milled and Polished

• 1
  Why the lenses of ASML lithography machines start as milled blanks
• 2
  Milling and diamond turning: geometry before figure
• 3
  Small grinding heads and the move to sub-surface damage control
• 4
  Magnetorheological polishing for figure and mid-spatial error
• 5
  Ion beam polishing and the final atomic-scale finish
• 6
  What a contract machine shop can and cannot do here
• 7
  Milling, grinding, MRF and ion beam compared

A projection lens for extreme ultraviolet lithography is not one piece of glass. It is a stack of mirror substrates and lens elements, and each one begins life as a blank that has to be shaped before any optical figure exists. The blank is generated by grinding and milling, not by casting to final form. That first shaping step decides how much material the later polishing stages must remove.

For reflective EUV optics, the substrate is usually a low-thermal-expansion glass-ceramic or a fused silica grade. It is brittle, abrasive to tooling, and expensive enough that a scrapped blank hurts. Shops shape these blanks on high-precision machine tools with diamond cup wheels and diamond ball-end tools, working wet to control heat and dust.

The milling step does not aim at optical accuracy. It aims at geometry: thickness, wedge, edge profile, mounting flats, and a surface close enough to spherical or aspheric form that the grinding head can take over with a predictable stock allowance. Think of it as roughing with a tolerance budget measured in micrometers, not nanometers.

That is the part of the chain a contract machine shop can realistically be asked to do. The optical figuring that follows belongs to a handful of specialist houses. Knowing where the boundary sits keeps a sourcing conversation honest.

• 1
  Blank materialGlass-ceramic or fused silica, milled wet with diamond tooling
• 2
  What milling controlsThickness, wedge, edge profile, mounting features, form allowance
• 3
  What milling does not controlOptical figure, surface roughness below Ra 0.2 μm, coating

On a 5-axis machining center, a diamond ball-end tool can generate a freeform surface by following a toolpath in three linear and two rotary axes. The rotary table keeps the tool normal to the surface, which spreads wear and avoids the rub marks you get when a ball tool drags at a shallow angle.

Typical parameters for glass-ceramic roughing sit in a narrow band. Depth of cut 0.02–0.1 mm, feed 50–200 mm/min, spindle 3,000–8,000 rpm depending on wheel diameter. Go harder and you seed subsurface cracks that show up two stages later as pits the polisher has to chase.

Coolant matters more than most people expect. A steady flood keeps the contact zone below the point where thermal expansion distorts the cut, and it flushes glass dust that would otherwise recirculate and scratch. Filtered coolant, not a sump full of swarf.

After milling, the part usually goes to a coordinate measuring machine for form and thickness. A ±0.005 mm envelope is realistic for the mechanical features. The optical surface itself is left with 20–100 μm of stock for grinding and polishing, depending on the curvature and the shop's process.

• 1
  ToolpathBall-end diamond tool kept normal to surface on a 5-axis center
• 2
  Roughing band0.02–0.1 mm depth of cut, 50–200 mm/min feed
• 3
  CoolantFiltered flood to hold thermal growth and flush abrasive dust
• 4
  Stock left for optics20–100 μm, set by curvature and the polisher's removal rate

Small grinding heads are rigid, pad-mounted abrasive tools, often 10–50 mm in diameter, spun on a precision spindle and driven along the surface by a CNC or a dedicated optical generator. Because the contact patch is small, the head follows curvature without needing a matching lap for every radius.

The job here is to remove the damaged layer left by milling. Fracture mechanics says the crack depth scales with abrasive grit size and load. So the shop steps down grit: perhaps 30 μm diamond, then 9 μm, then 3 μm. Each step must remove at least two to three times the damage depth of the previous one, or the deeper cracks survive.

This is where process discipline beats machine specifications. A shop that jumps from coarse to fine in one pass will measure a good surface and still fail a later acid etch test. The etch reveals subsurface damage that polishing only hides temporarily.

Grinding heads also set up the edge. Chamfers, bevels and mounting interfaces are finished here, because once the optical surface reaches nanometer roughness you do not want to touch it with a metal tool again.

• 1
  Head size10–50 mm pad-mounted abrasive on a precision spindle
• 2
  Grit sequenceRoughly 30 μm, 9 μm, 3 μm diamond with full damage removal each step
• 3
  Failure modeSkipping a grit size leaves subsurface cracks that etching exposes

Magnetorheological finishing uses a fluid whose viscosity rises sharply in a magnetic field. The fluid is carried on a rotating wheel, and a magnetic field stiffens it into a ribbon that behaves like a compliant, conformal grinding tool. Abrasive particles in the fluid do the cutting.

The useful property is that the removal rate depends on local shear. High spots see more shear and come down faster. That gives a deterministic, spot-based correction method. An interferometer measures the surface, software computes a dwell map, and the wheel removes material exactly where the figure is high.

MRF is good at correcting mid-spatial frequency errors, the ripples with periods of a few millimeters to a few centimeters that a full-size lap cannot reach. It is less effective at very long-wavelength figure error, where a larger tool is more efficient.

Removal rates are low, often a few micrometers per minute at the spot. That is fine, because by this stage only tens to hundreds of nanometers of error remain. The method is slow but predictable, and predictability is what an optics house is buying.

• 1
  MechanismMagnetically stiffened fluid ribbon on a rotating wheel
• 2
  Best atMid-spatial frequency ripples and deterministic spot correction
• 3
  Weak atLong-wavelength figure error, where a large lap is faster

Ion beam figuring removes material by sputtering. A beam of argon or another inert ion is rastered across the surface in vacuum, and the atomic layers are knocked off at a rate set by beam current, energy and dwell time. There is no contact, so there is no tool wear and no edge effect from a pad.

Because the process is non-contact, it can correct figure error on a surface that is already too good to touch. It also works on steep aspheres and on surfaces with features that a lap cannot reach. Removal is measured in nanometers per pass, with dwell maps computed from interferometer data.

The trade-off is cost and time. Vacuum chambers, ion sources and hours of raster time make this the most expensive finishing step. It is reserved for the last few nanometers of error, after MRF has done the bulk correction.

After ion beam figuring the surface is coated. For EUV mirrors that means a multilayer of molybdenum and silicon, tens of alternating layers, each a few nanometers thick. The coating preserves figure, so any error left before coating is locked in.

• 1
  MechanismInert ion sputtering in vacuum, non-contact removal
• 2
  Best atFinal nanometers of figure error on steep or finished surfaces
• 3
  Cost driverVacuum time, ion source hours, repeated interferometer metrology

The dividing line is straightforward. Milling and diamond turning of the blank, plus the mechanical features around it, sit inside the range of a well-equipped machine shop. Optical figuring below roughly Ra 0.2 μm, MRF, ion beam figuring and multilayer coating sit outside it.

That boundary matters when a program needs a mounting ring, a flexure, a housing or a metrology fixture for an optical assembly. These parts are machined to tight tolerances, often in low-expansion or high-stiffness materials, and they must not introduce stress into the optic when bolted up.

A shop working on these parts should hold ±0.005 mm on critical features, inspect 100% before shipment, and be able to show material certificates. Thermal stability of the fixture during machining is part of the tolerance stack, not a detail to be ignored.

GreatLight machines such parts on 127 high-precision CNC machines, including 16 simultaneous 5-axis centers and 16 mill-turn centers, with a maximum processing size of 4,000 mm. Materials include aluminium 6061 and 7075, stainless 304 and 17-4PH, titanium TC4, Inconel and engineering plastics. That covers the mechanical side of an optics build, not the optical side.

If a drawing asks for an aspheric surface at Ra 0.1 μm with a 2 nm figure error, say so early. The right answer is a specialist optics house, and the honest answer is that it is not a milling job.

• 1
  In scope hereBlanks, mounting rings, flexures, housings, metrology fixtures
• 2
  Out of scopeOptical figuring, MRF, ion beam figuring, multilayer coating
• 3
  Hold on critical features±0.005 mm with 100% inspection before shipment