Process guide

3D Printing Process: 7 Steps to Nanoglass Structures on Semiconductor Chips

This guide walks through a 3D printing process that writes quartz glass directly on a semiconductor chip surface. It is written for process and packaging engineers who need to judge whether a nanoglass route fits their part, which steps drive yield, and when a machined fixture is still the better answer.

Get Instant Quote Talk to an Engineer

Two-photon writing±0.005 mm fixturesRa 0.2–0.8 μm3–5 day parts

3D printing process writing nanoglass structures on semiconductor chips

On this page

Key takeaways

The glass forms in placeA focused femtosecond laser cures an organic-inorganic hybrid resin, then densifies it into quartz on the chip surface.

Substrate cleanliness decides yieldParticle and moisture control matters more than print speed at this scale.

Thermal budget is the hard limitDensification heat must stay under the metallization and passivation limits of the wafer.

Fixtures are still machinedWafer holders, chucks and alignment plates come off CNC, not off the printer.

Not every layer belongs in glassUse nanoglass for optical and dielectric features, and keep metal routing in metal.

What this process is

What the 3D Printing Process Actually Does

The 3D printing process covered here is not filament extrusion. A femtosecond laser is focused into a drop of organic-inorganic hybrid polymer resin sitting on the wafer. Two-photon absorption happens only at the focal point, so the resin crosslinks in a tiny volume instead of along the whole beam path. Move the stage and the focal point traces a line.

After writing, the unexposed resin is washed away with a solvent such as isopropanol or a propylene glycol ether. What remains is a printed structure roughly 70 to 90 percent organic by mass. A densification step follows, usually thermal or plasma assisted, which burns out the organic fraction and shrinks the part into amorphous quartz.

Feature sizes land in the sub-micrometer range, which is why the reported work targets nanoglass rather than bulk optics. Walls of 200 to 500 nm are routine in the literature, and aspect ratios of 3:1 to 5:1 are achievable before collapse becomes the limiting factor.

The reason this matters to chip designers is placement. The glass is written where the photonic or dielectric function is needed, directly on the die, instead of being fabricated separately, thinned, aligned and bonded. Every alignment step you remove is a yield loss you do not pay.

1
Resolution driverFocal spot size and stage stability, not resin viscosity.
2
ShrinkageExpect 15–30 percent linear shrink during densification; design compensation in.
3
Substrate limitWorks on silicon, glass and some compound wafers; not on soft or porous carriers.

Fit check

When a 3D Printing Process Fits and When It Does Not

Use this route when the feature is small, optically active, and needs to sit within a few micrometers of an existing device. Waveguides, microlenses, photonic couplers and low-k dielectric standoffs are the natural candidates. The value comes from co-locating glass with the circuit, not from the glass itself.

Skip it when the part is a structural housing, a heat sink, a connector shell or anything above roughly 1 mm in wall thickness. Densification time scales with volume, and thick glass sections crack during shrink. For those parts, machining or molding is cheaper and faster.

Skip it too when your wafer already carries low-temperature metals or polymers that cannot survive the densification window. If your backend stack is limited to 200 °C, a thermal densification step at 500 °C or above is off the table unless a plasma or UV-assisted variant is qualified for your stack.

A practical filter: if you can measure the feature with a caliper, it is probably a machining job. If you need an SEM to see it, the 3D printing process is worth a feasibility run.

1
Good fitSub-micrometer optical or dielectric features placed on-die.
2
Poor fitMillimeter-scale structures, load-bearing parts, high-volume simple geometry.
3
Blocked fitWafers with thermal budget below the densification temperature.

Setup

Resin, Substrate and Fixture Preparation

Resin choice sets your whole window. Organic-inorganic hybrid resins based on siloxane or zirconium-siloxane chemistry give the best conversion to quartz, but they are moisture sensitive. Store them cold, let them reach room temperature sealed before opening, and filter to 0.2 μm before dispensing.

Substrate prep is where most failures start. Spin or spray the resin to 5–50 μm, then bake briefly at 80–120 °C to drive off solvent. A 10-minute dehydration bake at 150–200 °C before coating removes adsorbed water and improves adhesion noticeably.

The wafer has to sit flat and stay flat. Vibration of 1 μm at the focal point blurs the write. This is the part we machine: vacuum chucks, wafer carriers and alignment plates flat to ±0.005 mm, with surfaces finished to Ra 0.2–0.8 μm so the wafer seats without rocking.

Keep the write area under a nitrogen purge if your resin is oxygen inhibited. Oxygen quenches the radical at the focal point and widens the effective voxel, which shows up as rounded, oversized features.

1
Coating thickness5–50 μm typical; thicker coats scatter more and cut depth of focus.
2
Pre-bake80–120 °C to remove solvent; avoid skinning the surface.
3
Fixture flatness±0.005 mm on the seating face keeps the focal plane stable across the wafer.

Execution

Step by Step: Running the 3D Printing Process

1
1. Clean and dehydrate the waferSolvent clean, then a dehydration bake at 150–200 °C for 10 minutes. Cool to room temperature before coating. Skipping the cool-down traps solvent under the resin and blisters the film later.
2
2. Coat the hybrid resinSpin or spray to 5–50 μm. Target ±2 percent thickness uniformity across the die area. A visible edge bead is normal; remove it with solvent before writing or it will catch the objective.
3
3. Mount on a machined chuckVacuum-clamp the wafer on a flat carrier (flatness ±0.005 mm). Check with a dial indicator that the surface moves less than 1 μm when you tap the stage. Any rock here becomes focus error.
4
4. Write with the femtosecond laserTypical parameters sit at 780–800 nm, 80–150 fs pulse width, 80–100 MHz repetition rate, and 1–20 mm/s scan speed. Raise power until features just develop cleanly, then back off 10 percent. Overpowering widens lines and burns the resin.
5
5. Develop the unexposed resinSoak in isopropanol or PGMEA for 2–10 minutes, then rinse. Do not scrub. Structures at 200–500 nm walls will shear off under any mechanical contact.
6
6. Densify to quartzThermal densification in air or oxygen, ramped slowly through the organic burnout window. Expect 15–30 percent linear shrink. A plasma or UV-assisted route runs cooler when the wafer cannot take the heat.
7
7. Inspect and measure shrinkSEM or optical profilometry for critical dimension, plus a refractive index check if the feature is optical. Compare against the pre-shrink model and adjust the CAD compensation for the next run.

Decision table

Nanoglass Printing vs. Machined Fixtures vs. Bonded Optics

Pick the column that matches the feature you actually need.

Criterion	3D printing process	CNC machined fixture	Bonded discrete optics
Feature scale	Sub-micrometer to a few μm	0.05 mm and up	0.1 mm and up
Placement	Directly on the die	Separate part, assembled	Separate part, aligned and glued
Alignment steps	None after writing	Fixture-to-wafer alignment	Optical alignment per part
Thermal limit	Densification window applies	Material dependent	Adhesive limit, often 150 °C
Wall thickness	Under 1 mm practical	Any	Any
Typical lead time	Feasibility run first	3–5 days after DFM	Depends on optic supplier
Best use	On-die photonic and dielectric features	Chucks, carriers, alignment plates	Standalone lenses and windows

FAQs

Common Questions

Can the 3D printing process replace lithography for these features?

No. It writes three-dimensional glass, not planar patterned films. Lithography is still better for high-volume planar layers with tight overlay. The 3D printing process wins when the geometry is genuinely 3D or when you cannot tolerate a separate bonding step.

Treat it as an addition to the flow, not a replacement.

What shrinkage should I design for?

Plan on 15–30 percent linear shrink from the printed green body to the densified quartz. The exact figure depends on resin chemistry and densification profile, so run a test coupon with your own recipe before committing a full layout.

Compensate in CAD by scaling the model up, then verify with a SEM measurement on the first article.

Does the laser damage the underlying circuit?

Two-photon absorption confines the energy to the focal volume, so the wafer beneath sees far less dose than the resin. Still, check your metal and passivation stack. Repeated exposure over the same metallization area can accumulate enough heat to shift a resistor value.

Run a split test with and without writing on a test wafer before running product.

How flat does the chuck need to be?

Flat enough that the focal plane does not drift out of the resin film. On our machined carriers we hold ±0.005 mm on the seating face and finish to Ra 0.2–0.8 μm, which keeps the wafer from rocking.

If you see focus drift across the wafer, measure the chuck first, not the optics.

Is this a production process today?

It is a qualified process for specific photonic and dielectric features, and it is still developing. Volume economics depend on how many wafers you write and how much alignment cost you remove.

For low-volume and prototyping work it is already practical. For very high volume, compare it against a dedicated lithography step.

Can you help with the machined parts around the printer?

Yes. We machine wafer chucks, carriers, alignment plates, brackets and housings to ±0.005 mm, with finishes from Ra 1.6–3.2 μm as-machined down to Ra 0.2–0.8 μm when the seating surface matters.

No minimum order quantity, from one prototype to 10,000+ parts, with quotation and free DFM analysis within 12 hours.