Outer Pixel Engineering Design Review Meeting
Mulvey Room
Oxford
-
-
10:00
→
18:00
Ladder production overview 8hSpeakers: Alex Knight (University of Oxford), Ashley McDougall (University of Oxford), Robert Becker (ETH Zurich)
-
13:00
→
14:00
Lunch 1h
-
10:00
→
18:00
-
-
11:00
→
13:00
Stiffener design review 2h
Adam (Brick Kiln Composites) will also join via zoom: https://cern.zoom.us/j/4528742764?pwd=HuTN2QQTIWGcus0e1oxIfm5n7Tvbn9.1
Topics of interest:
- Current design proposed by Adam/Alex. Status of prototypes? Any concerns with fabrication? Any concerns with stiffness? Where will the mechanical strain be forced through, the interposer flex? Okay with this?
- Other design styles we want to consider, e.g. extending all the way to the endpiece flex somehow, such that it's clamped down at the end.
- Vertex stiffener design from Robert?
Speakers: Alex Knight (University of Oxford), Ashley McDougall (University of Oxford), Matthew Brown, Robert Becker (ETH Zurich)Purpose
The meeting introduced Robert to BrickKiln’s composite stiffener technology, developed for the Mu3e outer pixel ladders. The team wanted to learn from BrickKiln’s experience for a more modular vertex detector design, which was exploratory. Lessons learnt from the outer pixel stiffener development and experience (from the last ~ 2.5 years) was discussed.
Stiffener design and history
- Principle: stiffness comes from section depth/shape, not extra material.
- Original design:
- It used a single ply of 22 µm unidirectional carbon fibre, custom-specified through a Korean supplier. The material has since been discontinued.
- A 8 µm Kapton layer was co-cured on the back and doubled as a cooling flap sealing against the adjacent ladder. 3 µm Kapton exists but costs much more.
- It ended before the chip edge, which left a weak spot between the stiffener and the interposer. Other problems with the current design: limited access to probing points, difficulty reworking bonds.
- Noise problem: testing found parasitic capacitive coupling, with the stiffener picking up noise in the chips. The stiffener was flipped from the chip side to the reverse side of the ladder, which fixed it.
- New design:
- It shifts the stiffener to give chip access and runs it onto the interposer, with a flat foot on each interposer tab for bonding.
- Three geometry variants exist: two smaller lumps, a single raised section, and a lower single curve with a larger bond area.
- A thicker or woven material over the interposer was rejected. The transition from thin to thick material would add material over the chips, unidirectional fibre can’t change shape much without breaking, and it would need different tooling.
- Hybrid carbon-Kevlar “metamaterial”:
- Kevlar sits where the stiffener touches the chip, with carbon on the raised “apex” sections.
- It removed the capacitive coupling, and its cured ply is about 12 µm.
- Adam said it is proven, not theoretical. Kapton would no longer be needed for chip isolation, though it is still wanted as the sealing flap and for handling robustness.
- The resin system used for this material was proprietary, something developed by BrickKiln with the company directly, therefor the details won’t be shared. This applies also to the actual “splicing in” process that they use to make this meta material.
Manufacturing- The process is now very repeatable but delicate. The production run of 40 (L3 2026) had one failure, against roughly 4 made per good part in early attempts.
- Problems solved along the way:
- Wrinkling on the bag face. A silicone intensifier shrank over time and was replaced by a proprietary rubber, which may no longer be needed.
- Resin shrinkage, which bowed early prototypes by about 20 mm. The fix was confirmed to carry over to the new shape.
- Breaking parts off the tool, now done with an air cushion.
- Adam said the process used to extract the parts from the mold tool are trade secrets and he can’t share them. This was generally true about all aspects of the manufacturing procedure done at BrickKiln, mainly motivated by the fact that they use similar processes for other more secretive projects.
Key discussion points- Kapton cost and role: Kapton is the largest cost driver. Adam said it also adds handleability and robustness, so he’d want a mechanical-ladder test before removing it.
- Extending the stiffener over the full interposer:
- Extending the current profile in length is easy, because BrickKiln already makes them oversize and trims them. Matt later added that changing the end-piece clamp to have a u profile would be very easy.
- Changing the cross-section along the interposer would be real R&D, and a thicker or different material would also cover the bonds.
- Matt is keen to extend the stiffener all the way out into the clamp region. Decision to be taken once the current prototypes are received and can be tested.
- The conclusion on any significant R&D work: not possible for next year (2027); fine to explore for the future.
- Vertex concept:
- A thick carbon end frame around the interposer, with the thin stiffener glued across the ladder (triangle shape with feet to glue to ladder - chip side?).
- Feasibility of flexure-based ladder mounting (Z-profile springs) to decouple thermal expansion from ring structure — noted as future concept, not scoped for next year
- The vertex detector had failures when it was clamped to rigid frames, and the cause is still being analysed (mismatch stresses are suspected).
- For 2027 the design is essentially locked.
- Feedback on Robert’s early V-shaped concept (prototyped in his earlier CMS work):
- Adam advised a curved apex rather than a sharp one, because sharp edges in unidirectional fibre split the resin.
- He was worried about bond area and about how the V sections would be attached to the ladder, since a stiffener that isn’t attached well does nothing.
- He also found the transition to the end frames hard to judge from the screen.
- Adam/Alex suggested swapping in BrickKiln’s smooth curve and adding a bonding lip.
- Resin:
- Robert asked about cyanide ester, which is common for stability and radiation reasons. Adam said the metamaterial uses a proprietary resin, not epoxy.
- Adam would be happy to develop it with a standard cyanate ester if that gives shared knowledge.
- Adhesives: Adam said radiation-hard structural adhesives qualified for the LHCb VELO exist (an adhesive transcribed as “CAF4” was mentioned, he wants to double-check the grade).
Collaboration, IP and commercial terms- The Mu3e side is uneasy about relying on a proprietary process or resin, because technology intended for wider use at CERN would need radiation qualification, plus also the general ideas of sharing knowledge across projects/DRDs etc. They can run their own testing at CERN and share results openly.
- Adam is happy to give informal feedback and do some R&D at BrickKiln’s own cost and risk, as it is already doing for the redevelopment of the new stiffener. He needs an indication of where it’s going commercially, such as future orders, to justify the work internally. He noted there may be more orders for the existing stiffeners possible.
- BrickKiln still has the tooling for the original vertex stiffeners and could quickly try different materials or geometries.
- The preference on the room side is to adapt the existing profile, add their own frames, and use BrickKiln for production runs, where yield know-how matters most.
- Joey said the 2027 plan should be clear within about two months. Larger changes would be a 2028 topic.
Status updates- Prototypes of the new design: 3D-printed tooling produced flat parts (the main thing Adam wanted to check), but the tools are porous and left the stiffeners resin-dry. They were made without Kapton. Adam didn’t know whether the batch due last week had been finished.
- Vertex stiffeners shipped to Joey: received after a customs and payment delay. Joey hasn’t tested them because he is the only mechanics person on vertex.
- Material availability: the Skyflex stock at Oxford is past its expiry date, and a DMA test to confirm its properties is pending. BrickKiln have plenty more of the existing carbon-fibre material that can be used. More Kapton can be ordered quickly from Goodfellows and doesn’t expire, but it is expensive, so Adam suggested not ordering until the prototyping is finalised.
Action itemsAdam:
- Check Skyflex data sheet for resin system
- Oxford resin is K50 (epoxy)
- We have H23 (cyanate ester) and K50 and K51 (a slightly higher temp epoxy with a different cure)
- Update later how latest prototype batch came out
- Send radiation-qualified adhesive grade (CAF4) to Robert
- (Alex) waiting on ordering more Kapton material until prototyping stage is completed
Joey/Ashley
- Update Adam with 2027 production intentions once known (after early Dec collaboration decision)
Robert
- When ready, send Vertex design files to Adam for informal feedback and possible prototyping
Oxford MEG (via Alex)
- Complete the DMA test on the expired material
-
13:00
→
14:00
Lunch 1h
-
14:00
→
17:00
Module design discussions 3h
Topics of interest:
- End-ring redesign to ensure connectivity
- End ring alignment
- Module assembly: any modifications needed?
- Mounting in QC box, any improvements to be made?
- Cooling okay?
- Cabling & connectors -reorientation?
- Plans for 2027, timeline for redesigns? (Largely depends on collaboration goals which will be decided in December, but we can discuss options).
Meeting organisation: what will be discussed between "production"/"installation"/Robert's engineering meeting?
Speakers: Ashley McDougall (University of Oxford), Matthew Brown, Robert Becker (ETH Zurich)1. L3 module alignment and pins
- What happened in 2026:
- The pin-and-slot design held the pin rigidly in a slot allowing only horizontal movement, which caused most of the problems in the QC box.
- On the cage, the pins were removed and modules were installed with no guidance, by eye. This worked surprisingly well. Good connections were seen on all but one half module (M6 DS).
- An L3 service flex was damaged by an LV short, so a non-working half module was placed in that position.
- Proposed fix for 2027:
- Remove the closed end of the slot, so it guides the module IN, without preventing vertical movement. Modifying existing end pieces is a small machining job. Plan to test this with the next module in Liverpool within the next few weeks.
- It can be fully tested in the QC box before installation, and if it doesn’t work the pins can come out again.
- There are no interposer alignment pins, and none are possible.
- The modules need to come apart anyway, so trying this is low extra risk.
2. L4 10x20 interposer clamp- A low-profile clamp aligns and connects the 10x20 interposer to the end piece and service flex. It uses FR4 boards, PEEK countersunk bolts to align, and glued nuts to retain them.
- L3 installation couldn’t go as planned in 2026, because of an end-piece flex collision and removal of the mounting locations. The clamp is the alternative connection. It still needs manufacturing and testing.
3. End ring redesign- Problems with the old design: L3 and L4 share one ring, with many separate helium exhausts and 90-degree hydra connections. Alignment is insufficient, and the ring has to be installed first.
- New concept:
- Separate L3 and L4 rings, each split in half so they can be installed after the sci-fi.
- A compliant layer clamps around the beam pipe to seal against backflow.
- Helium distribution is combined into split, 3D-printed manifolds.
- The L3 ring keeps the 2026 alignment cutouts and ears, and the L4 ring drops the interposer locations. Overall thickness is unchanged.
- Installation order: L3 ring after sci-fi, then L3 modules, then L4 ring, then L4 modules.
- Open design questions:
- How the two ring halves align and bolt together (radially), and how the rings mount to the beam pipe.
- The beam-pipe brackets probably need to change.
- Vertex and sci-fi cooling pipes are not yet considered.
- Whether the rings can be aluminium (the room’s view: probably, with insulation behind the flexes, which Kapton tape could cover).
- How to insulate the two halves to stop current circulating, and spring loading (two separate rings would allow separate springs).
- Other ideas:
- Reverse the helium flow direction, which was set aside because warm air would have nowhere to go.
- Use the downstream beam pipe for flow in future. Matt called this possible but not for now.
- Slot LV cables through the rings rather than under them, to ease the spring-load resistance.
- Spring loading and the QC box: the aim for 2027 is to make the rings as free as possible and allow some motion for thermal expansion, since 2026 ended up over-constrained. The more flexible mount is longer-term.
- Mock-ups: the existing mock-ups were not realistic enough for cabling. Plans are to test on the real cage as early as possible, make better mock-ups and spare real rings, and start the CAD model fresh (Robert plan) from the latest versions (it contains many old versions).
4. Hydra, cooling and flow- Hydras: bundles of cables blocked the hydra routing. Hydras aren’t rigid enough for 90-degree connections, and two of the four hydra clamp locations were impossible to connect. The fix is straight flexible ducts instead of the elbow, with the aluminium part redone and the old 3D-printed pieces removed.
- Backflow: a quick fix from Robert was used in 2026, but a permanent one is needed. Options are blocking the cable gaps in the existing ring, or the new ring that seals against the beam pipe.
- L4 flow: L3 flow was good, but L4 is a concern. The plan is to test it, ideally on the cage next year, with Thomas’s findings as input. Thomas is on leave from December, so he’ll be asked by email soon (by Matt). Questions remain on how to quantify and visualise flow.
- Outer foil around L4: there are gaps between ladders where helium escapes into the magnet bore. The room wasn’t convinced a foil helps and could even make things worse, so it should be modelled with and without. This can run in parallel with other work.
5. L3 power and cabling- Litz-type cable: a very flexible multi-strand silicone cable. Sample bought at a hobby shop, spec sheet not available. It’s much more flexible than the current cables, but needs a current rating check.
- Idea: solder flying leads directly to the module-flex ears, with a cable-to-cable inline connector, so LV power doesn’t go through the 10x20 interposer or the service flex. Fewer screw connections is a goal. A solder-blocking liquid could stop solder wicking up and stiffening the cable.
- Scope: changing the service flex itself is too big for 2027. Connectors would only go where accessible (downstream is easier).
- Service flex orientation: doing the same for L3 would need the module flexes unfolded, and there isn’t space to hide it next year. It’s flagged as a long-term goal for all layers.
6. QC box- Problem: each module has to be mechanically and electrically installed in the QC box, which is unreliable, time-consuming, and in practice only Matt can do it.
- Proposal: keep modules in their transport/storage boxes during QC, so only the electrical connection is made, using the clamp. Feasible for L4 and uncertain for L3, because the flex would have to be unbent. It would be quicker, safer, and more people could do it. A service flex would always be attached and ready.
- Cooling the box:
- The air currently comes in at about 15°C but chips still reach 50 to 60°C.
- Ideas: a cold channel or heat-pipe duct feeding the box, cooled walls, or nozzles along the ladder.
- A vortex tube went to about −30°C in a test but with very low flow, so there is a trade-off between flow and temperature.
- Thermal bowing: after a long cold run, the ring frame stays cold while the ladder warms and expands, which bows the ladder. The easiest fix is a slow temperature ramp-down (longer test time). Other ideas were a carbon-fibre base or a springed mount. Everyone agreed it isn’t worth much time but should be watched.
- Ladder transport: cracks were seen after transport to Liverpool (two chips, not end chips). They were packed tightly together. The fix is individual foam cut-outs like the module transport case, which means more cases or more trips.
7. Planning, meetings and resources- 2027 production and promises:
- A statement is needed in December on whether the systems can produce a physics detector.
- The room would not promise a full detector. The line was “we build what we can, carefully”, with a layer 4 target and some layer 3 replacement ladders.
- The aim is 95%+ efficient modules, and wouldn’t be wise to rush at the expense of yield learning.
- Production approach:
- Build modules often rather than in a batch, for example one a month.
- Test the first few ladders in a module before committing to a large run.
- Keep ladder production going unless modules show a ladder problem.
- Investigate failures.
- Chip supply:
- About 65 wafers are left, and L4 alone needs roughly 37 to 40 wafers at current yield.
- Full recurl stations aren’t possible, and a new chip is maybe years away (P2Pix chip isn’t working yet … ).
- Ladder yield was around 60%, which burns chips quickly.
- Resources:
- About 28 ladders are needed for L4, plus about 12 for L3 replacements (roughly 40).
- Ladder production is about three mornings a week on the gantry, shared with ATLAS ring work, so around 20 weeks.
- Alex is the only person making ladders. Two part-time chip testers are expected soon and the new student could help with QC. Technicians (Liverpool) could help with module building or installation.
- UK funding becomes clearer in spring.
- HDI and QC supplies: more HDIs need ordering. The new flex HDI is still being tested, with hoped-for fixes to the main yield problem (SVTAP1 issues).
- Chip yield issues: chips sometimes fail for unknown reasons, and broken chips and dust are a continuing problem.
- Meeting structure:
- Too many overlapping meetings (production, installation, integration, engineering, HDI), with the same topics repeated.
- Direction: cut rather than add, and move off Friday if needed.
- Keep a straightforward production meeting for component flow, yields and QC, and move design changes to an engineering meeting.
- The engineering meeting would run every other week and could alternate with another meeting, possibly on Mondays.
- A periodic “kick-off” or in-person meeting (roughly three days, around spring) was proposed to discuss the longer-term layout and the full detector, after beam time.
-
11:00
→
13:00