3D Printed Pulmonary Tissue for Toxic Smoke Testing

Ammonia and chlorine move through ports and factories in large volumes every day. When a release happens, the first question is always the same: what does this concentration do to a human lung, and how quickly? Animal tests are slow, expensive and hard to defend. A flat petri dish tells you almost nothing about how a gas behaves at an air-liquid interface. That gap is why labs build a living airway model instead.

A lung-on-chip puts human pulmonary cells on a thin membrane and feeds air past one side, culture medium past the other. The cells stretch and flex the way they do in a breathing lung. Dose the air stream with a controlled gas or smoke mixture, and you can watch the tissue respond in real time. The chip is small, so you can run many concentrations in parallel on one bench.

The tissue itself is printed, not machined. Cells are laid down layer by layer into a pattern that matches alveolar or bronchial geometry. That part belongs to the biology lab. The housing that holds the membrane, seals the gas path and lets a microscope see through must be made to tight tolerances, and that is where CNC work enters the picture.

A typical chip is the size of a glass slide, roughly 25 × 50 mm, sometimes smaller. Inside, microchannels run at 60–200 μm wide. The top layer carries the air path, the bottom layer carries the medium. A porous membrane sits between them, and the pulmonary tissue grows on that membrane. Two ports feed gas, two feed liquid.

The channel geometry sets the flow profile. At 60 μm the pressure drop is steep, so surface finish inside the channel matters more than most people expect. A burr of 20 μm can change the flow and shift your dose curve. That is the reason we machine these channels with small end mills and then inspect them under a toolmaker's microscope rather than trusting a visual check.

Optical access drives the material choice. If the lab images through the chip, the top and bottom plates must be clear. If the chip only needs to be functional, opaque plastic is fine and cheaper. We have cut both, and the trade-off is almost always clarity versus cost, not clarity versus accuracy.

• 1
  Channel width60–200 μm typical; smaller channels need finer tooling and slower feed rates
• 2
  Membrane area10–50 mm² of active tissue; larger areas need better sealing
• 3
  Port countTwo gas, two liquid is common; extra ports raise the sealing risk
• 4
  Wall thickness1.5–3 mm gives enough stiffness without blocking the optics

Most chip housings are machined from PMMA, PC or COC. PMMA gives the best optical clarity and cuts cleanly, but it is brittle and chips at the edges if you push the feed too hard. PC is tougher and handles sterilization better, at the cost of a slightly hazier view. COC sits between the two and is a common pick for molded versions, though machined COC blanks are harder to source in small quantities.

The channels are the hard part. A 200 μm channel with a 3:1 depth ratio is routine on a 3-axis mill with a 100–200 μm cutter. Below 100 μm, tool breakage climbs fast, so we usually cut the plate in two halves and bond them rather than try to reach the full depth in one pass. That way each half is an open channel, easy to inspect and easy to clean before bonding.

Flatness across the sealing face decides whether the chip leaks. We hold 0.01 mm over the whole plate on the mating surface, and we check it on a granite surface plate before it leaves the shop. A warped plate will not seal against a 60 μm channel no matter how good the gasket is.

For the fixture that holds the chip during a smoke test, the demands are different. It has to seal against gas, hold the chip flat and let a probe or camera reach the ports. We machine those from 6061-T6 or 316L depending on whether the lab autoclaves them. Stainless survives repeated steam cycles; aluminum is lighter and cheaper but needs anodizing if the surface will see condensation.

Machining wins for one-off chips, prototype geometries and fixtures. If you are testing a new channel layout every week, a machined plate lets you change the drawing and cut a new one in a few days. No mold, no tooling cost. We regularly run single chip bodies and small fixture sets for lab groups that are still iterating on the design.

Injection molding wins once the geometry is frozen and you need hundreds of identical chips. A molded chip has better surface finish in the channel and a lower unit cost at volume, but the mold takes weeks and the design cannot change without a new tool. Most programs run machined parts through the research phase and switch to molding only when the protocol is locked.

There are cases where CNC is the wrong answer. If your channel is 20 μm wide, neither machining nor molding will hold it reliably, and you are better off with a glass or silicon process. If the chip needs embedded electrodes on a 10 μm pitch, that is a semiconductor job, not a machine shop job. Say so early; it saves everyone a month.

The same logic applies to the smoke delivery side. A gas mixing manifold with a few ports is a straightforward machining job. A manifold that has to hold a calibrated 0–500 ppm ammonia stream within 2 percent is an instrument, and it should be built by someone who does flow calibration, not just metal removal.

Not every dimension on a chip plate needs to be tight. The outer profile can sit at ±0.1 mm and no one will notice. The channel width, the sealing face flatness and the port positions are the three that change results. We quote those separately so you can see where the cost is.

Channel width we hold at ±0.005 mm when the cutter allows it. Sealing face flatness we hold at 0.01 mm across the plate. Port positions we hold at ±0.05 mm, because a misaligned port puts a step in the flow path and creates a dead volume where smoke can linger.

Surface finish inside the channel sits at Ra 0.8–1.6 μm after a light finishing pass. Going finer than that rarely helps flow and adds cost. Going coarser than Ra 3.2 μm starts to trap cells and residue, which means the chip is harder to clean and reuse.

Everything ships with a dimensional report if you ask. For medical and lab work we hold ISO 13485 as well as ISO 9001, and we can sign an NDA before you send drawings. Uploads stay confidential whether or not an NDA is in place.