Motionwell Automation builds cleanroom material handling equipment in Singapore, and wafer handling automation is the corner of that work where the substrate rather than the mechanism dictates every decision. The delivered record behind it is electronics: a series of five cleanroom automated test equipment variants, each carrying 4 to 6 fan filter units that hold ISO Class 7/8 over the fixtures, one of which is a cleanroom-compatible electronic tester for semiconductor devices with automated docking, thermal cycling and data logging; collaborative-robot loading holding plus or minus 0.05 mm at the test fixture interface, calibrated against a reference pin at the start of every batch; vision-guided placement on an ABB SCARA sensor panel line at plus or minus 0.01 mm; and wafer transfer machines for cleanroom-compatible transport between process stations, built on precision servo mechanisms with ESD-safe handling. Machines are designed, assembled and tested at our Woodlands Link facility, and the company has delivered more than 150 special purpose machines since 2014 under ISO 9001:2015 and bizSAFE Level 3.
Where we stand, said plainly before you read further, because vagueness here is expensive for the buyer. The sentence above about wafer transfer machines is the whole of our published record on that work: no throughput figure, no particle count, no reliability number, and none invented to win a page position. We do not build process chambers, load ports, sorters or equipment front end modules, we do not sell an EFEM as a product, and we are not a notified body. What we design and build is the machine between and around process tools: the frame, the motion, the tooling, the enclosure and its airflow, the control system, the recovery logic and the record. Where a proven standard tool from a semiconductor equipment supplier covers what you need, that is the cheaper and faster answer and we will say so at concept review rather than quote around it.
This page is therefore written as engineering rather than as a portfolio. It covers why contamination and breakage come before motion, what the carrier settles before anything is drawn, how the end effector families divide, why the handoff tolerance is the one that matters, what a particle class does to drives and cables, why static is a design input, what a broken wafer costs after it breaks, where the seconds go, what a per-wafer record carries, and when a catalogue tool beats anything we would build. The general cleanroom engineering is on our cleanroom automation equipment page and is not repeated here. If you already have a substrate, a carrier and a station list, skip ahead and talk to an engineer.
Why Is Wafer Handling a Contamination Problem Before It Is a Motion Problem?
Because the motion is the easy part and everybody underestimates how easy. A wafer moves a short distance, in a plane, slowly, with almost no mass, along a path that never changes. What is being bought is a machine that performs that trivial move a very large number of times without touching the wrong surface, without shedding, without charging the disc, without cracking it, and without needing a person to reach inside.
The economics explain the asymmetry. A wafer arriving at a mid-line transfer step carries the accumulated cost of every process step already performed on it, and it carries that cost across its whole active surface. A machine that drops it destroys all of that at once, and a machine that lands a particle on it may destroy some of it silently, with the loss appearing at electrical test weeks later against a cause nobody can now identify. That is the same silent failure signature we design around on electronics test work.
That sets the order the machine gets designed in. The contamination boundary comes first, because it fixes the enclosure, the airflow and where the mechanism may live. The contact scheme comes second, because it fixes the end effector and with it the acceleration budget. The mechanism is drawn third, against those two. Reversing the order produces a fast handler that cannot hold its class, and the fix is a new frame.
What Do the Carrier and Its Interfaces Settle Before the Machine Is Drawn?
More than the robot choice does. The carrier decides where the wafer meets room air, and that single answer sets the size and cost of the clean envelope.
| What the material arrives in | What it is | What the machine has to provide |
|---|---|---|
| Front opening unified pod (FOUP) | A sealed carrier, opened only at a load port, so wafers never see room air in transit | A load port with door handling, a mapping sensor, and a clean envelope that only has to cover the transfer path |
| Sealed pod with a bottom-opening interface (SMIF) | The same principle in a different geometry for smaller formats | The same door-handling problem, resolved against a different opening direction |
| Open cassette | A slotted carrier with no door, so the wafers breathe whatever air surrounds them | A clean envelope over the whole machine, not just over the transfer path |
| Carrier delivered by overhead or floor transport | A drop-off at a defined position rather than a hand-carried box | A load position with presence and orientation sensing, and an interlocked handshake before either side moves |
| Handoff signalling between transport and load port | The parallel I/O interlock defined by SEMI E84 | Correct sequencing and, more importantly, a named owner for the signals |
| Equipment-to-host messaging | SECS/GEM, the standard route for equipment state, events and per-unit data | A control architecture that can report state as events rather than as a screen |
Two points from that table are where projects actually go wrong.
Slot mapping happens before motion or it happens too late. A cross-slotted or double-slotted wafer that is discovered by a blade already entering the carrier is a broken wafer and possibly two. The map is taken first, the machine refuses to start on an ambiguous map, and the refusal is a feature that operators will ask you to remove in month two.
Interfaces need an owner named in writing. Every row above is a boundary between two suppliers, and each is a place where both can reasonably assume the other holds it. The cheapest place to fix that is the specification, which is what our guide to writing an automation URS is for.
Which End Effector Holds a Thin Brittle Disc, and What Does Each Cost?
The choice is decided by which surface is allowed to be touched, not by preference or by cost. Once that is answered, the rest follows.
| End effector | How it holds | What it touches | What it costs you |
|---|---|---|---|
| Passive support blade | Gravity and friction alone, on a flat paddle | The back surface, over the whole pad footprint | Acceleration is capped by friction, and there is no grip signal to verify, so a slip is discovered by its consequences |
| Vacuum blade | A sealed area under the back surface, held by pressure difference | The back surface, with the largest contact print of the three | Needs a flat, clean, sealable back face. A leak is not a weak grip, it is no grip, so the vacuum level has to be monitored rather than commanded |
| Edge grip | Actuated contact on the wafer edge only, clamping laterally | The edge exclusion zone, with both faces clear | An actuator, its utilities and its wear on the blade, a wider swing envelope, and a much tighter requirement on where the wafer sits before the fingers close |
| Bernoulli | A gas flow across the front face producing a low-pressure region that lifts without contact | Nothing, or edge stops only | A continuous supply of clean dry gas, high flow moving directly over the product surface, and lateral instability unless the wafer is constrained |
The Bernoulli row deserves its own sentence because it is the one that gets specified for the wrong reason. Non-contact is not the same as low risk. The tool works by moving a lot of gas immediately above the active surface, which means the cleanliness of that gas and of everything the flow can reach becomes product quality. A non-contact effector in a machine whose airflow was never designed for it can move more contamination onto a wafer than a clean vacuum pad would.
Whichever family wins, blade materials are chosen for the environment as much as for stiffness: engineering plastics such as PEEK and technical ceramics behave inside a clean zone, while painted or powder-coated parts are a particle source that worsens over the machine’s life. The wider treatment of tooling design, including how a grip is verified rather than commanded, is on our end of arm tooling page.
Why Is the Handoff Tolerance Tighter Than the Transport Tolerance?
Because they are two different requirements that get written as one number. Transport tolerance is a clearance question: the blade has to enter and leave a slot without scraping, and the wafer has to stay on the blade. Handoff tolerance is a capture question: the wafer has to land inside the acceptance range of a chuck, a pocket or a set of lift pins, and that range is far smaller than the slot clearance.
The number people quote is robot repeatability, and it is one term in a chain that also holds teach point accuracy, blade deflection under load, where the wafer actually sat on the blade, carrier slot position, the station datum, and thermal growth between a cold start and hour six. Repeatability is usually the smallest of them. Budgeting the chain rather than quoting the arm is what makes a handoff as reliable at hour six as at hour one.
Orientation is the second half of the problem and it is not solved by position. Downstream steps need the wafer’s rotational orientation known, and the wafer does not remember it. An aligner rotates the disc under a sensor, finds the notch, or the primary flat on older formats, along with the centre offset, and reports a correction the robot applies on the next move. Two consequences follow. It is a serial step unless the layout buffers it or a second arm covers it. And it is a measurement, so it also reports how far off-centre the wafer arrived, which is a free early warning about a drifting station that most machines discard.
This is the same architecture as the vision compensation on our sensor panel line, where the camera measures the actual offset in X, Y and theta every cycle and the robot trajectory shifts, instead of a mechanical nest being made for every variant. The exposure that architecture creates should be stated plainly, because it applies identically to a notch: the machine depends on the feature being present, clean and unobstructed, and a damaged feature does not degrade gracefully, it stops. That trade is set out in full in the vision-guided SCARA panel assembly case study, and the sensing side of it on our machine vision inspection page.
What Does a Particle Class Do to the Drives and the Cables?
It moves them. The mechanism sheds as it wears, and on a handler it sits directly above the product for the whole stroke rather than beside it, so it goes below the product plane, behind a seal, or both, with pneumatic exhaust ducted away rather than vented over the wafer. Those choices, and the material, finish and lubricant decisions attached to them, are the same for any machine holding a class and are covered in the cleanroom machine design guide.
Two consequences are specific to this work and worth adding here.
Moving the servo drives outside the enclosure, which is the standard answer for a particle source with its own cooling fan, buys longer shielded motor cables. In a machine that also has a static control requirement, those cables are where grounding and electromagnetic compatibility start to argue with each other: the shield wants one defined path to ground, the ESD scheme wants every accessible surface bonded, and resolving both without creating a ground loop is a design task at schematic stage rather than a wiring decision on the shop floor.
And the airflow has to survive the machine’s own motion. A blade sweeping across a wafer drags air over the product surface, so the downflow path over the transfer route has to be genuinely clear. That constraint fixes where frame members, cable trays and overhead tooling can sit, which makes it a layout decision rather than a detail.
Why Is Static a Design Input Rather Than an Accessory?
Because charge is generated by the machine doing exactly what it was built to do. A wafer lifting off a blade is separation. A carrier door opening is separation. A belt turning, a plastic guide rubbing, dry filtered air moving fast over a surface: all generate charge, all are designed in, and all happen long after a bill of materials was signed with “ESD-safe materials” written on it.
The useful design question is not which materials are safe but which electrical role each surface has to play, because the answers conflict.
| Electrical class | Delivered example on our machines | The role it plays | What goes wrong when it is used in the wrong place |
|---|---|---|---|
| Insulative | Nylon and POM specimen nests on the cleanroom test equipment, surface resistivity exceeding 10^12 ohm | Keeps the fixture from loading or shorting the device during resistance and continuity measurement | An insulator cannot bleed a charge. Rub it and it holds what it collects until something else provides the path |
| Dissipative | Carbon-fibre-loaded PVC conveyor belting at 10^6 to 10^9 ohm, dissipative classification per IEC 61340-5-1 | Bleeds charge to ground gradually rather than in one event | Useless without a real path to ground behind it, which is the part that gets value-engineered out |
| Grounded and bonded | Conveyor frames bonded to the facility ground bus with straps at every 3 metre interval on our battery line | The return path every other measure silently assumes exists | A bonded frame with an insulating coating between the bond and the working surface is decoration |
| Ionisation | Grounded workstations with ionisers on our electronics handling work | Neutralises charge on things that cannot be grounded, including the product itself | An unbalanced or unmaintained ioniser is a charge source, so it needs a verification routine and not just an installation date |
The first two rows together make the point that a materials list cannot. A test fixture wants insulation so the measurement stays honest, and the handling path wants dissipation so charge cannot accumulate. Both are correct, they are not the same part, and deciding which is which surface by surface is engineering rather than procurement. That conflict is invisible on a specification asking only for ESD-safe construction, which is why we treat the static scheme as a drawing rather than a line item. The electronics context for that work sits on our electronics and semiconductor automation page.
What Happens After a Wafer Breaks?
It will break eventually, and what the machine does next decides whether the cost is one wafer or a shift.
Fragments go where the geometry and the airflow send them: into the slot, along the blade, down into whatever is under the transfer plane, and onto the wafers still in the carrier. That is why a machine that stops in place and holds its state is worth more than one that completes the commanded move. The questions somebody needs answered immediately are which wafer, which slot, whether anything is still on the blade, whether the rest of the lot is suspect, and whether the machine can be opened without dragging debris further into the clean envelope.
Designing for that answer costs little at concept and cannot be added later. A defined safe stop that does not finish the move. Blade and slot geometry a person can actually see into with a torch. Service access from the non-clean side wherever the mechanism allows, with a documented re-clean where a panel must open into the clean zone. Captured fasteners, so nothing else joins the debris during recovery. And an event record holding the last commanded position and the last confirmed sensor state, because the machine will be cleaned before anyone gets round to diagnosing it.
Restart is the part that gets skipped. A machine that resumes by trusting its memory of where it was is not safe after a breakage. It has to re-map the carrier, re-verify presence at every station it believes is occupied, and require a positive human acknowledgement before it moves. That sequence is written once, at design time, or it is improvised at three in the morning by whoever is on shift.
Recovery is also the safety case rather than an exception to it, because it is the scenario that puts a person inside a machine with servo axes and stored pneumatic energy. Where a robot is inside the enclosure, the robot and the integrated system fall under ISO 10218-1, with each safety function carrying a required performance level under ISO 13849-1, calculated rather than asserted. The guarding and validation scope that follows is on our machine safety and CE marking page. The figure worth asking any supplier for is not a mean time between failures but how long a recovery takes and who is qualified to perform it, because inside a clean envelope the recovery includes the re-clean.
Where Do the Seconds Actually Go on a Transfer Machine?
Not where the datasheet suggests. The arithmetic is simple and it is almost never done before a robot is chosen.
| Element of the cycle | What it is | Does it overlap the process |
|---|---|---|
| Carrier dock, door open, slot map | Fixed overhead once per lot | No, and it is charged to every lot regardless of size |
| Station to carrier move, unloading the finished wafer | Per wafer | No. The station is idle throughout |
| Carrier to station move, loading the next wafer | Per wafer | No. The station is still idle |
| Alignment | Per wafer | Only if it is buffered or handled by a second arm |
| Settling and hold verification before release | Per wafer | No, and it is the step most often left out of an estimate |
| Process time at the station | Per wafer | This is the number the whole machine exists to protect |
| Assists and exception handling | Unpredictable | This is why the measured rate differs from the calculated one |
A station’s output is set by whichever is longer: the process itself, or the handling round trip that has to fit between two processes. If the round trip fits inside process time, the tool is the constraint and a faster robot buys nothing. If it does not, the fixes run in order of cost: overlap the exchange so the outgoing wafer is held while the incoming one is placed, which is what a second blade or a second arm is for; move alignment off the critical path; add a buffer; and only then move faster. Speed is the last lever because acceleration on a brittle disc is capped by the hold rather than by the drive.
The number that decides your real output is assists, not cycle time. A machine that indexes beautifully and needs a person twice a shift produces less than a slower one that does not, and the difference does not appear in any specification. Ask for a mean time between assists during acceptance, and write the measurement method into the protocol before the machine is built. The same argument on a multi-station tester is worked through on our automated test equipment page.
What Does a Per-Wafer Record Have to Carry?
The unit of record is the wafer rather than the lot or the carrier, and that decision belongs in the control architecture rather than in a reporting layer bolted on afterwards.
A useful record binds an identity to a place and a history: which wafer, which carrier and slot it came from and returned to, which station and which arm handled it, timestamps at each transfer, the alignment correction that was applied, and every exception including the ones the machine recovered from on its own. That last pair is where the value sits. A correction that drifts on one station over three weeks is a mechanical fault reporting itself long before it becomes a broken wafer, and a machine that applies the correction without logging it has thrown that warning away.
The architecture is one we have delivered on a different unit of record. On the cleanroom test equipment, pallet identification uses RFID tags embedded in the pallet body so the PLC tracks each specimen’s position and test status around the loop, force, displacement, resistance and continuity are logged per serial number, and export runs as CSV by batch for offline Cp/Cpk study or over OPC-UA for live charting on Siemens WinCC or equivalent. Substitute the wafer for the specimen and the carrier slot for the pallet, and the structure is unchanged.
One caution decides whether any of it is worth keeping. A record is only as good as its identity binding. If the machine infers which wafer it is holding from a slot position, and anything re-orders the carrier between two operations, every downstream record is confidently wrong and nothing flags it. Bind to a read wherever the substrate carries a readable mark; where it does not, the binding lives in the carrier and the slot, and the machine has to stop on the ambiguous case instead of guessing. How that binding is designed and verified generally is on our code reading and traceability page.
When Is a Catalogue Tool the Better Answer?
Where the work is standard, which is more often than a custom builder’s website usually admits. A standard wafer in a standard carrier, moving between standard load ports, being sorted, transferred or inspected, is a mature product category with established vendors, regional spares and a large installed base. Nothing drawn from scratch competes on price, lead time or the odds of working in week one.
Custom work earns its place at the edges of that description, and the edges are specific. The substrate is not a standard wafer, so it is a panel, a carrier tray, a frame of singulated die or a non-round part nobody sells a handler for. The transport runs between tools never designed to talk to each other, which makes the interface the project rather than the motion. The machine has to do something as well as move, meaning test, measure, mark or cure, which is the shape of our delivered electronics work. A working legacy tool has no usable interface, so the real choice is a bridge or a replacement, covered on the legacy machine connectivity page. Or the binding constraint is the building rather than the process. How to run that decision is set out on our custom machine versus standard equipment page.
What Has to Be Settled Before Anyone Draws the Machine?
Seven questions, in this order. The first that gives a hard answer usually reshapes everything after it.
- What is the substrate, and which surfaces may be touched? This settles the end effector family, and with it the acceleration budget.
- What does it arrive in, and where does it meet room air? Sealed carrier or open cassette decides whether the clean envelope covers a path or a machine.
- What is the station’s capture range at the handoff? Not the robot’s repeatability, but the capture range and the tolerance chain that has to fit inside it.
- Does anything downstream need known orientation? If it does, alignment is a station in the layout and a step in the cycle.
- What is the process time at each station? Handling only has to fit inside it, and everything faster is money spent on nothing.
- What has to be true after a breakage? Where fragments go, who opens the machine, what gets re-cleaned, and what restart demands.
- What must be provable per wafer, and to which system? Decide it before the control architecture, because it is expensive to retrofit.
| Your situation | Start from | Why |
|---|---|---|
| Standard wafers, standard carriers, sorting or transfer between load ports | A catalogue tool from a semiconductor equipment supplier | A mature product beats a drawing on price, lead time and installed base |
| Non-standard substrate, or a carrier nobody sells a handler for | A purpose-built machine | There is no catalogue answer to compare against |
| Both faces of the substrate are process surfaces | Edge grip, or Bernoulli with edge stops | A back-surface pad is not available, and that decides the tool before the robot |
| The wafer is landing outside a chuck’s capture range | The tolerance chain and the teach points, not a more repeatable arm | Repeatability is usually the smallest term in the error budget |
| Handling time already fits inside process time | Overlap, buffering and reliability work | A faster robot buys nothing when the tool is the constraint |
| The line runs, but somebody attends it twice a shift | Assist rate and recovery design | Sustained output is set by interventions rather than by cycle time |
| A working legacy tool with no usable interface | A connectivity bridge assessed against replacement | The interface is the project; the motion is the easy half |
Inter-station transport across a plant, rather than at one tool, sits on our warehouse and intralogistics page.
Which standard editions apply right now?
The editions below are the ones we design and document against on current projects. We check them on the date shown rather than assuming last year's edition still holds.
| Standard | Current edition | What it means for your machine |
|---|---|---|
| ISO 10218-1 — Robotics, safety requirements, Part 1: industrial robots | ISO 10218-1:2025 | Published February 2025, the third edition and the first substantive revision since 2011. It adds robot classifications with matching functional safety requirements, safety-related cybersecurity requirements, and end-effector guidance. Most of ISO/TS 15066:2016 on collaborative operation moved into Part 2. |
| ISO 13849-1 — Safety of machinery, safety-related parts of control systems | ISO 13849-1:2023 | The 2023 edition is the version referenced by ISO 10218-1:2025 for robot control system safety functions. Designs still documented against the 2015 edition will need their PL calculations restated when the machine is re-assessed. |
Frequently Asked Questions
Has Motionwell built a wafer handling machine?
We build wafer transfer machines for cleanroom-compatible movement between process stations, using precision servo mechanisms and ESD-safe handling, and that sentence is the full extent of what we will claim. We hold no published throughput, particle or reliability figures for that work, so we quote none. Our documented depth in this sector is electronics rather than semiconductor front end: five cleanroom automated test equipment variants holding ISO Class 7/8 over the fixtures, one of them a cleanroom-compatible tester for semiconductor devices with automated docking, thermal cycling and data logging. Process chambers, load ports and equipment front end modules are a different industry and we do not make them.
Can a substrate handling machine be changed over to a different wafer size later?
Assume not without significant rework, and price the alternative while the layout is still on paper. Carrier interface, slot pitch, blade geometry, edge-grip finger positions, aligner chuck, teach points and the whole tolerance chain are sized around one substrate. What normally survives a size change is the frame, the enclosure, the control architecture and the software. What does not survive is everything that touches the wafer. Where two sizes are genuinely in the plan, say so at concept: designing for both from the start changes the envelope of the machine, and it costs a fraction of retrofitting the second one afterwards.
What can a factory acceptance test actually prove about a cleanroom handling machine?
Motion, sequencing, interlocks, recovery behaviour and the data record, all of which are worth witnessing in person. What it cannot prove is the particle performance you will live with, because that figure only means something when it is measured under motion at production speed in your own environment, which puts it in performance qualification rather than at handover. Decide the sampling points and the acceptance criteria during concept design, because they change where the access ports go. Machines are built and tested at our Woodlands Link facility on a 16 to 24 week lead time from concept approval, so a Singapore buyer attends that test rather than flying to it.