The problem, stated concretely #
We needed a hollow structure that mimicked a bone-marrow canal: 15 mm outer diameter, 8 mm inner, 30 mm long. It had to end up around 65 ± 3 % porous, with those pores interconnected — isolated voids are useless, because cells and nutrients have to travel through the structure, not sit in sealed pockets. And it had to survive being handled and loaded.
The obvious instinct for a 2020s engineer is to print it. You cannot, at least not easily. A hydroxyapatite powder is not a thermoplastic; there is no filament to melt. To extrude it you must first suspend it in something, and the moment you do that you are fighting two requirements that pull in opposite directions: enough ceramic in the mix to survive firing without collapsing, and enough fluidity to actually pass through a nozzle.
So the route we took avoids the problem entirely: print the negative, not the part.
What we actually did #
flowchart TD
A["CAD negative: hollow half-cylinder
15 mm OD / 8 mm ID / 30 mm"] --> B["FDM-print PLA mould"]
B --> C["Cast slurry:
60% Mg/Si/Zn-HAp in PEG + PVA
+ methylcellulose anti-settling"]
C --> D["Dry 120 °C / 48 h"]
D --> E["Burn out organics 600 °C
PLA mould + binder removed"]
E --> F["Sinter 1200 °C / 2 h
ceramic densifies"]
F --> G["Helium pycnometry
65 ± 3% interconnected porosity"]
F --> H["Uniaxial compression
18 ± 2 MPa"]
A PLA mould is FDM-printed as the inverse of the part. It gets filled with a slurry that is 60 % Mg/Si/Zn-HAp powder, suspended in polyethylene glycol and polyvinyl alcohol with methylcellulose added to stop the ceramic settling out before it dries. Then three thermal stages, and the distinction between them is the single most misread part of this process:
| Stage | Temperature | What happens |
|---|---|---|
| Drying | 120 °C, 48 h | Water leaves. Slow, because fast drying cracks the green body. |
| Debinding / burnout | 600 °C | The PLA mould and the polymer binder burn away. What remains is a fragile ceramic skeleton. |
| Sintering | 1200 °C, 2 h | Ceramic particles fuse. The part gains its strength here. |
Burnout and sintering are not the same step, and they do not happen at the same temperature. I have seen this collapsed into a single “fired at 1200 °C” in more than one summary of our own work, and it matters: everything organic must be gone before densification starts. If the binder is still decomposing while the surface is closing up, the escaping gas has nowhere to go and the part cracks from the inside. The 600 °C hold is what buys you a clean part; the 1200 °C hold is what buys you strength. All the parameters are in the data sheet.
The result: 65 ± 3 % interconnected porosity measured by helium pycnometry, and 18 ± 2 MPa in uniaxial compression.
Why not just print the ceramic directly? #
Because as of today, nothing in the published literature reaches 18 ± 2 MPa at 65 ± 3 % porosity by direct extrusion. That is the honest answer, and the table below is the evidence for it rather than a tour of other people’s work.
For scale: human cancellous bone runs about 2–12 MPa in compression (Budharaju et al., 2023), and its Young’s modulus spans 1.3–22.3 GPa (Wu et al., via Hassan et al., 2025).
| This work — indirect sintered | Si-doped BCP — direct extrusion | Sr/Zn-HAp/collagen/PLGA — electrospun | SLA / DLP — vat photopolymerisation | |
|---|---|---|---|---|
| Method | CAD negative mould, slurry cast, sintered | Robocasting through a 400 µm nozzle | High-voltage electrostatic spinning | Photocurable ceramic-loaded resin |
| Compressive strength | 18 ± 2 MPa | >2 MPa at ≤4 mol% Si | Not evaluated in bulk | DLP-HAp 15.25 MPa; DLP ZrO₂/HA 39.99–52.25 MPa; SLA resin 5 MPa |
| Young’s modulus | Not reported | Not reported | 9.91 ± 1.7 GPa | — |
| Porosity | 65 ± 3 %, interconnected | 70–80 % open | High, but pores 2.9 ± 0.76 µm | CAD-defined |
| Pore scale | Macro, interconnected | 400–500 µm square lattice | Sub-micron — below what cells need | 100–1000 µm |
| Source | This paper | Lu et al., 2024 | Hassan et al., 2025 | Budharaju et al., 2023 (review) |
Four things in that table are worth stating plainly, because summaries of this literature get them wrong often enough that I want the corrected versions on the record:
Neither this work nor Lu et al. measured Young’s modulus. Both report compressive strength. Those are different quantities and a scaffold comparison that fills in a modulus column for studies that never measured one is inventing data. Only the electrospun work has a real modulus, and it was obtained by nanoindentation, not bulk compression — which is again a different thing.
Lu et al.’s silicon threshold is ≤4 mol%, not >4 mol%. Scaffolds doped at 4 mol% silicon or below held 70–80 % open porosity and compressive strength above 2 MPa. It was the 8 and 10 mol% formulations that dropped below 2 MPa. The doping level buys you biological activity and costs you strength, and the interesting region is the boundary — not a blanket “more silicon is worse”.
Vat photopolymerisation covers a wide range, and quoting the top of it is misleading. In Budharaju et al.’s survey the actual reported numbers are DLP hydroxyapatite at 15.25 MPa, a DLP ZrO₂/HA composite at 39.99–52.25 MPa, and an SLA acrylic resin at 5 MPa. A separate lithography-based route (LCM, alumina, 50 % porosity, 500–1000 µm pores, 48 h sinter) reached a compression Young’s modulus of 2–5.5 GPa with fractural strength of 11–133.5 MPa. That 2–5.5 GPa figure belongs to LCM alumina specifically — it is not a general property of SLA/DLP.
Electrospinning’s weakness is geometric, not mechanical. Hassan et al. produced beautiful 331.6 ± 96.11 nm fibres, but the resulting pores were 2.9 ± 0.76 µm. Their own paper points out the problem: osteoblasts are 10–50 µm and need pores of 100–200 µm to migrate into a structure. A mat with excellent stiffness that cells cannot enter is a surface, not a scaffold.
The process engineering nobody puts in the abstract #
Three things dominate whether a ceramic part comes out of the furnace usable, and none of them are chemistry.
Shrinkage is enormous and you must design for it. Across the cases Budharaju et al. survey, sintering shrinkage runs from about 12 % to 45 %. A 20 wt% HA/TCP ink shrank 45.25 %; raising the ceramic loading to 40 wt% brought it down to 17.8 % at 1150 °C. Infill density does the same thing from the other direction — 100 % infill gave 12 ± 3.7 % shrinkage, 75 % gave 28 ± 3.8 %, and 50 % gave 42 ± 4.7 %. If your CAD model is the size you want the finished part to be, you have already failed.
And the shrinkage is not uniform. Green bodies shrink more in Z than in XY. The mechanism is weak binding between deposited layers and easier binder escape along that axis, compounded by isotropic sintering stress — not, as I have seen claimed, gravity or settling. Which means the compensation factor is anisotropic: you scale Z differently from XY, and you have to characterise it empirically for your own slurry and furnace profile.
Nozzle geometry sets your floor on pore size. Lu et al. extruded through a 400 µm inner diameter nozzle to produce a lattice with 400–500 µm square through-holes. That is comfortably above the 100–200 µm cells need, and well past the >150 µm that Picado-Tejero et al. (2025) associate with vascularisation and osteogenesis. Their review also finds that gyroid and diamond TPMS geometries distribute compressive load better than square or star lattices — a genuinely mechatronic result, since it says the toolpath, not just the material, determines the mechanics.
Where I would take this next #
This is my own view as the engineer on the team, not a claim from the paper, and I’ve ordered it by distance from what we can actually do now.
First, slurry rheology. The reason we cast rather than printed is that we did not have a formulation that could be extruded and still survive firing. That is a solvable, unglamorous problem: raise the solid loading, get shear-thinning behaviour so the paste flows under the nozzle and stops the moment it leaves, and pick binders that burn out cleanly. It is squarely a mechatronics-and-process problem rather than a chemistry one, and it removes an entire manufacturing stage — no mould to print, no mould to burn out.
Then simulation before fabrication. Finite-element analysis of the stress field and computational fluid dynamics of nutrient perfusion through the pore network, run on the CAD model before anything is fired. Firing a batch takes days; simulating a geometry takes minutes.
Then higher resolution, via SLA or DLP with ceramic-loaded resins, where the anatomical geometries that matter clinically — cranial, maxillofacial — become reachable.
Then multi-material, a stiff ceramic lattice carrying load with a softer phase delivering growth factors.
And eventually, cell-laden bioprinting. I want to be blunt that this is the horizon and not the next step. Going from a sintered acellular ceramic to printing live cells is not one increment; it is a different manufacturing paradigm, and anyone who presents it as the obvious follow-on hasn’t costed the intervening work.
Worth stating clearly: the paper’s own recommended next steps are different from mine. It calls for in vivo animal models, biomechanical evaluation under cyclic loading, growth-factor loading, and optimisation of the doping ratio. Those are the priorities of the biologists and chemists on the team, and they’re right for the material. Mine are the priorities of the person who had to make the object.
What this study does not establish about manufacturing #
Scoped deliberately to fabrication — the biological limits are covered in the closed-loop breakdown and listed in full on the data sheet.
- Mechanical evaluation was static compression only. No fatigue testing, no cyclic loading. A scaffold that survives one load is not a scaffold that survives a million, and human bone gets loaded on the order of a million times a year.
- The scaffold’s properties are partly an artefact of the indirect route itself. A directly extruded part from the same powder would not necessarily behave the same way.
- One geometry, at one scale, from one furnace profile. Nothing here establishes how the process behaves across a size range or on a different machine.
- The paper limits its claim to non-weight-bearing defects, and so do I.
Frequently asked #
What is indirect 3D printing of ceramics?
What is the difference between debinding and sintering?
How much do ceramic parts shrink during sintering?
Why not use direct ink writing instead of a mould?
What pore size does a bone scaffold need?
Was this scaffold tested in a living body?
My role on this project was systems translation — turning chemical requirements into something that could actually be manufactured and measured. I’m actively looking for bioinformatics and biomedical engineering collaborations where that’s the missing piece: groups with strong chemistry or biology who need CAD, systems architecture, process design and manufacturing execution to get from a formulation to a physical object. If that’s your lab, get in touch.
