What custom microfluidic device fabrication costs

The short answer: a custom single-layer PDMS microfluidic device costs $1,500 of one-time tooling plus $80 per chip at the one fabricator we found publishing a rate card, which works out to $1,580 for one part and $95 per part at a hundred. An SU-8 master mould on its own is $790 to $1,150 commercially or $1,060 to $3,295 at an academic cleanroom.

If an off-the-shelf device fits your experiment, it will beat all of that: eNUVIO’s OMEGA-MP is $440 for 4 devices covering 8 experiments, which is $55 per experiment with no tooling cost at all.

The two numbers that make up every quote

Every microfluidic fabrication quote decomposes into non-recurring engineering (NRE) and per-part cost.

NRE is design, photomask, master mould, tooling and process qualification. You pay it once, regardless of how many parts you order. It is the number that determines whether a prototype run is affordable.

Per-part is material, cycle time and handling. It is the number that determines whether production is affordable.

The mistake buyers make is optimising the wrong one. At n = 1, NRE is 95 percent of the bill and per-part cost is irrelevant. At n = 1,000, NRE has almost vanished and per-part cost is everything. Choosing a fabrication method without knowing which regime you are in is how programmes end up paying injection-moulding tooling costs for a run of forty devices.

The published pricing

Volume tiers: the arithmetic

Taking HiComp’s published $1,500 NRE plus $80 per chip and amortising it:

With a chrome photomask added at $500, so $2,000 of NRE:

Two observations from that table.

Per-part cost falls by 95 percent between n = 1 and n = 100, then essentially stops falling. Between 100 and 1,000 parts the per-chip cost moves from $95 to $81.50 — a 14 percent improvement for a ten-fold increase in commitment. Soft lithography does not scale, because every chip is an individual casting operation. There is no volume at which it becomes cheap.

The crossover to a different method sits somewhere above 100 parts. uFluidix publishes 100 chips at $60 each, so $6,000 with tooling apparently included, against $9,500 for the same quantity on HiComp’s structure. That is a real 37 percent difference at n = 100 and it is the kind of spread that makes running two RFQs worthwhile.

If you already have a master: mould amortisation only

Buying a master mould and casting PDMS yourself is the academic default. Here is what the mould alone contributes per part, excluding PDMS, labour, plasma bonding and yield loss:

This looks unbeatable and it is not, because the excluded costs are the majority of the true per-part cost. A PDMS casting cycle — weighing, mixing, degassing, pouring, curing, peeling, punching, cleaning, plasma treating, aligning, bonding — is several hours of elapsed time with roughly one to two hours hands-on per batch. At the $55 per hour academic staff rate published by the same facility that sells the $1,060 master, an hour and a half of hands-on time is $82.50 per batch, which dominates the mould amortisation at every tier above n = 10.

The honest reading: buying a master is cheap. Casting from it is not, and the cost is labour, not materials. If your technician’s time has any real cost, the $80-per-chip outsourced route is competitive far earlier than the mould price suggests.

The number that most affects this calculation is your yield. Bonding failures, leaks at inlet ports and alignment errors are the common modes. We could not find a published failure rate from any fabricator, so we do not assert one — but a 20 percent loss rate raises your effective per-part cost by 25 percent, and that alone can move the buy-versus-make answer.

Method comparison

Where we have a published figure, we say so. Where we only have a vendor’s own published analysis rather than a price list, we mark it as an estimate. Where we found nothing, we say nothing.

Two vendor analyses circulate widely and are worth reading with the right label attached:

• Parallel Fluidics publishes an analysis stating that each injection-mould design revision requires new tooling at $100,000 or more with lead times of 6 weeks or longer, that programmes typically need five to seven tooling cycles rather than the one or two they budget for, and that total development cost to a manufacturable device runs $100,000 to $1 million. This is a vendor’s own framing of a problem their product addresses. It is credible directionally and it is not a price list.
• Eden Microfluidics publishes a detailed cost breakdown covering raw material cost per chip by material, cleanroom access rates, SU-8 mould costs, plasma bonder capital cost and worked annual scenarios. Again: a vendor’s analysis, published to make a case for their own material and equipment, useful as a structural map of where cost hides, not as a source for a number you would put in a budget.

We are not reproducing figures from either as if they were market prices. The distinction between a rate card and a vendor’s cost narrative is exactly the distinction this site exists to maintain.

The off-the-shelf comparison, which usually wins

Before commissioning anything, price the catalogue devices. eNUVIO publishes live prices for its PDMS culture device range:

$55 per experiment with zero tooling. Against a custom route at $1,580 for the first part, the off-the-shelf device buys you 28 experiments before you have spent what one custom part costs.

This is the price ceiling every custom quote has to justify itself against. The only reasons to go custom are geometry the catalogue does not offer, a feature size the catalogue does not hold, or a volume high enough that per-part cost dominates. “We would prefer our own design” is not one of them at n = 10.

Note also that eNUVIO’s higher-throughput 96-format devices return no public price — the store API returns zero for those SKUs, meaning quote-only. The published prices are for the low-throughput formats.

What actually drives a custom quote up

From the published specifications and constraints of the fabricators above:

• Layer count. Penn QNF prices a single-layer master at $1,060 and a two-layer master at $3,295 — a 3.1× multiplier for one extra layer. Every alignment step is a large step change, not an increment.
• Channel height. Blue BioDevice’s pricing tiers on height alone: $790 at 5 µm and above, rising to $1,150 at 300 µm and above. A 46 percent increase across the range.
• Feature size and tolerance. Penn QNF’s standard process covers feature widths above 20 µm within ±10 percent, and heights up to 100 µm within ±20 percent. Anything tighter is explicitly “subject to additional review” and additional fee.
• Aspect ratio. Both fabricators cap at roughly 1:4 width to height as standard. Tall narrow structures are prone to breakage and delamination.
• Design maturity. Penn QNF bills design assistance and any consultation beyond the first 30-minute meeting at $55 per hour, charges for DXF-to-GDS conversion at the same rate, and states that any change after fabrication begins incurs charges for processing already executed. Arriving with a finished GDS file is worth real money.
• Iteration count. This is the largest hidden cost in every custom programme. Each design revision in a soft lithography workflow means a new master. At $790 to $3,295 per master, five iterations is $4,000 to $16,500 before a single usable device exists.

The decision rule

1. Check the catalogue first. At $35 to $141 per experiment with no tooling, an off-the-shelf device beats custom fabrication for any programme under roughly 30 devices unless the geometry is genuinely unavailable.
2. Below ~50 parts, minimise NRE, not per-part cost. Soft lithography, single layer, standard feature sizes, an existing photomask if possible.
3. Between ~50 and ~500 parts, run at least two RFQs. The published spread at n = 100 between $6,000 and $9,500 is 37 percent and it is free to discover.
4. Above ~1,000 parts, soft lithography stops making sense — per-part cost has flattened at roughly $80 and will not improve. This is where tooling investment starts to be rational, and where the tooling-iteration risk described by Parallel Fluidics becomes the dominant programme risk rather than a footnote.
5. Freeze the design before you commission anything. Every fabricator in this survey charges for changes after work starts, and the master mould is the item you re-buy.

What to put in a fabrication RFQ

Quotes are only comparable if they answer the same question. Specify:

• File format, ideally GDS. DXF conversion is billable.
• Minimum feature width and the tolerance you actually need, distinguished from the tolerance you would like.
• Channel height and tolerance, since height drives the master price tier directly.
• Number of layers, and whether alignment between layers is required.
• Material: PDMS, thermoplastic, glass. This changes the method, not just the price.
• Substrate and bonding requirement — bonded to glass, to another PDMS layer, or supplied unbonded.
• Surface treatment or coating, if any.
• Sterilisation requirement, if any.
• Quantity now, and realistic quantity within twelve months. Quote both.
• Whether the master mould is delivered to you or retained by the fabricator. This determines whether your second order is a per-part order or a full NRE again, and it is the single most commonly omitted line in an RFQ.

That last point deserves emphasis. Blue BioDevice sells you the master, so you own the tooling. Others quote a chip price with the tooling retained. Two quotes that look similar at n = 100 can diverge completely at your second order, and the difference is a contractual term rather than a price.

How we can help

Microfluidic fabrication quotes are hard to compare because fabricators bundle NRE differently, some sell you the tooling and some keep it, and almost none publish per-part pricing at volume.

We put the same specification to multiple fabricators, normalise the responses into the NRE plus per-part structure used above, and check who retains the master. Where an off-the-shelf device would do the job — and for culture applications it frequently would — we will tell you that rather than run a custom quote cycle.

Send us a specification.