Organ on a Chip Cartridge Tooling

• 1
  Key takeaways
• 2
  What an organ on a chip cartridge tooling job actually controls
• 3
  Channel geometry: where micromilling works and where it stops
• 4
  Material choice drives both biology and machinability
• 5
  Surface finish, bonding, and the assembly stack
• 6
  When to machine, when to mold, and how to tell
• 7
  Process fit by feature size and volume

A cartridge is not a flat plate with grooves. It carries a fluid network, a porous membrane, at least one culture chamber, and often a mechanical or optical interface. The tooling decides four things that no downstream process can recover: channel geometry, wall finish, material chemistry at the wetted surface, and how the membrane seats.

Take the membrane seat. A porous PET or PDMS membrane is typically 10–50 μm thick and has to sit flat against a ledge without crushing. The ledge height and the channel depth have to land within a few microns of each other, or the membrane sags into the channel and the cross-section changes. Flow rate then drifts between cartridges, and the drift is invisible until you run a perfusion experiment.

The same logic applies to the inlet ports. A luer or barb fitting that sits 0.05 mm proud creates a dead volume where cells settle. A port that sits 0.05 mm recessed leaks at the seal. Both parts pass a dimensional inspection and both fail in use.

So organ on a chip cartridge tooling is mostly a stack-up problem. Each feature is easy on its own. The difficulty is holding the relationship between features across multiple setups, and proving you held it.

Micromilling handles a lot of cartridge work well. Typical channels run 100–500 μm wide and 50–200 μm deep. With a 200 μm two-flute carbide cutter at 30,000–40,000 rpm and 200–600 mm/min feed, you can hold ±0.005 mm on channel width in aluminum or a rigid thermoplastic. Floor finish lands around Ra 0.4 μm with a clean finishing pass.

Deep, narrow channels are a different story. A 100 μm wide channel cut 500 μm deep is a 5:1 aspect ratio. The cutter deflects, the wall tapers, and chips pack into the slot. Taper of 10–20 μm over the depth is common and it is not a machine error. It is tool deflection and chip evacuation.

The practical ceiling for straight micromilling is around 5:1 to 8:1 depending on material. Beyond that, options narrow. You can use a tapered cutter and accept a trapezoidal cross-section, which is often fine for flow but changes shear stress at the wall. You can step the depth in multiple passes with a smaller depth of cut. You can switch to a laminated stack where each layer is a simple pocket, then bond the layers. Each route has a cost.

For channels under 50 μm wide, micromilling is usually the wrong answer. Photolithography on silicon or SU-8, hot embossing, or injection molding from a lithographic master will give better sidewall verticality. If your design needs 20 μm channels, say so early. It changes the whole quote.

• 1
  Straight walls100–500 μm wide, up to 5:1 depth ratio, ±0.005 mm achievable
• 2
  Tapered cuttersPast 5:1, accept a trapezoid and recalculate wall shear
• 3
  Laminated stackBest for deep features; adds a bonding step and alignment tolerance
• 4
  Sub-50 μmLithography or embossing from a master, not micromilling

PDMS is still the default for prototyping. It is gas permeable, optically clear, and easy to cast from a machined or lithographic master. It is also soft. A PDMS cartridge deforms under clamp pressure, and the deformation changes channel height. For a master that will be reused to cast dozens of parts, PDMS is fine. For a production mold insert, it is not.

Rigid thermoplastics are the production route. Cyclic olefin copolymer (COC) and COP give good optical clarity and low autofluorescence, which matters for imaging. Medical-grade polycarbonate is stiffer and cheaper but has higher autofluorescence. PMMA machines cleanly and polishes well, though it is brittle and can craze with some solvents.

Machinability ranking is roughly POM, PMMA, PC, COC, then PEEK. COC is notch sensitive and can chip at thin edges. PEEK is machinable but abrasive and expensive, and it needs sharp tooling and generous coolant. If a supplier quotes COC and PEEK at the same price, ask what tooling they plan to use.

For metals, 316L stainless and titanium grade 5 are common for reusable manifolds and mold inserts. Both are gummy. 316L work-hardens quickly, so light radial engagement and constant feed matter more than spindle speed. Aluminum 6061 is the fast option for fixtures and alignment plates, but it is not a wetted material unless it is coated.

One more constraint: any cutting fluid, release agent, or mold release that touches a wetted surface has to be validated for cytotoxicity. This is where general machine shops fall down. The part is dimensionally perfect and biologically unusable.

Channel wall finish does two jobs. It sets optical clarity through the wall, and it sets wettability. A polished floor around Ra 0.2–0.8 μm gives good imaging and predictable filling. An as-machined floor at Ra 1.6–3.2 μm traps air in the corners and makes priming slower.

Bonding is where most cartridge programs lose yield. Thermal bonding of COC or PC needs a temperature and pressure window that closes the channel without collapsing it. A 100 μm deep channel can lose 10 μm of height in a slightly hot press, and the loss is not visible until you measure flow. Solvent bonding is faster but leaves residue. Plasma activation followed by thermal bonding gives a strong, clean seal but the activation decays, so the window between treatment and bonding is measured in minutes.

Alignment between layers is a tolerance stack, not a single number. If layer A has a ±0.02 mm feature position and layer B has the same, the bonded assembly can be off by ±0.04 mm. For a 200 μm channel feeding a 300 μm port, that is workable. For a 50 μm channel feeding a 60 μm port, it is not. Design the alignment features with slack before you design the channels.

Fixtures belong in the same conversation. If the cartridge is imaged under a microscope, the fixture has to hold it flat to within the depth of field. That is often tighter than the channel tolerance, and it is the part people forget to quote.

The decision usually comes down to stage and volume. Micromilled cartridges make sense for 1 to a few hundred units, for design iterations, and for any geometry that is still moving. Turnaround is days, not weeks, and a design change costs a new program rather than a new mold.

Injection molding makes sense when the geometry is frozen and you need thousands of parts with identical optical and surface properties. A mold insert itself is a precision machining job, often with features that are harder than the final part because the insert is steel or a hard copper alloy. Tool steel, 17-4PH, and beryllium copper are common insert materials. Beryllium copper conducts heat better and cycles faster, but it is softer and wears at thin ribs.

Hot embossing sits between the two. A machined or lithographic master presses into a thermoplastic sheet. It handles sub-50 μm features better than milling and needs less pressure than injection molding, but cycle time is longer and the master wears.

A useful test: if the smallest feature is wider than 100 μm and the aspect ratio is under 5:1, micromilling is probably cost-effective at any volume up to a few thousand. If features are under 50 μm, or the aspect ratio is over 8:1, plan for a lithographic master and a replication step. Everything in between is a judgment call, and it is worth asking for both quotes.