| Episodes | 146 |
| Mentions | 426 |
| Cited here | 69 |
| First — last | #6 — #723 |
| Top guests | Mike Harrison, Joe Garrison, Chris Denney |
| Related | reflow oven · contract manufacturer · pcb assembly · pcb fabrication · led |
A pick-and-place machine is an automated assembly machine that lifts surface-mount electronic components from feeders with a vacuum nozzle and sets them onto a printed circuit board coated with solder paste.[411] It forms the central station of a surface-mount assembly line, a sequence of physically separate machines joined by a conveyor comprising solder-paste application, pick and place, a reflow oven, and automated optical inspection, with the board entering at one end and emerging assembled at the other.[50] The machine matters economically because the alternative is hand placement at roughly three hundred components per hour, a rate any placement machine must at least match to justify itself.[317][224]
Operation
In the line, the stencil printer first aligns stencil apertures to the board and prints solder paste, a slurry of tacky flux and powdered solder; the pick-and-place machine then sets components into the paste, whose tackiness holds them in position until the reflow oven melts the solder.[411] In a well-run shop the only manual steps are loading boards at the head of the conveyor, an inspector between placement and reflow who looks for catastrophic faults without touching the board, and through-hole insertion, which has resisted automation while functional test is normally fixtured and automatic.[121]
Because the board travels the line on a conveyor, a production panel needs a strip along its outer edge carrying sprocket holes, which the conveyor rails and the bed-of-nails tester use to advance and align it.[646] Test coupons and manufacturing strips added at the top and bottom of a panel carry these holes; they are a design-for-manufacture feature rather than an electrical requirement.[494]
Design-for-manufacture review for an assembled board checks clearance between parts for the placement head and whether heavy parts sit on the side that must pass through the process twice; at quantities of ten or twenty a shop can jig around such faults, while at five hundred the review governs whether the board must be laid out again.[716]
The machine itself has no way of knowing whether the reel loaded into a slot is the part it was told to expect; a wrong-value reel is caught only downstream at automated optical inspection, which compares each board against a photographed golden reference and can read component markings.[554]
Feeders
Feeders, not the motion system, are what make a machine a pick and place: a gantry with a vacuum head and no tape feeders cannot run commodity parts continuously and is therefore no faster than hand placement.[224] The cost of a placement machine is concentrated in the feeding mechanism and the vision system rather than the XY motion, which is why the class of machine that reliably takes components off reels starts around ten thousand dollars.[49] Feeders are typically the most expensive consumable part of a line, on the order of five hundred dollars each, and an assembly house wants as many as possible because a part already sitting in a feeder removes setup time from every job that uses it.[317]
Feeder count sets the number of passes a board needs; if the machine cannot hold every unique part at once, the board must be run through a second or third time, which removes the machine’s time advantage over hand placement.[63] A typical machine carries between twenty and forty reel positions, so the number of unique parts on a board, not the number of placements, determines setup cost, and adding one part past the slot count forces a second setup.[216] A machine populated with sixty feeders cannot place a sixty-one-part board in a single pass, so a bill of materials listing more uniques than the machine has slots costs an extra pass even when the extra part is trivial.[508] Unconsolidated passive values consume feeder capacity quickly: a board carrying around sixty distinct resistor types would fill a typical machine with resistors alone.[500]
These economics shape design practice. Designers reduce the unique-part count on a bill of materials specifically to stay within the assembler’s feeder capacity, because keeping a board at twenty reels instead of twenty-one avoids a changeover.[141] An integrated part that replaces a discrete circuit is bought partly for feeder economics: one reel on the machine instead of four is a direct reduction in setup and slot pressure.[580] Reusing a higher-rated part already loaded on the line rather than specifying the exactly correct lower-rated one is normal practice, because a second reel occupying a feeder slot for one component costs more than the part difference.[574]
Newer feeders accept loose cut tape, so a customer can order the exact quantity of a part from a distributor rather than a full reel and the assembler does not have to re-reel it.[184] Short cut-tape strips carry a leader, however, and a machine will consume a substantial fraction of a ten-part strip before it places reliably, so buying exact quantities of expensive parts for machine placement can waste most of them.[410] An assembler must in any case be supplied with more components than the board count requires, because winding a reel onto the machine loses parts at the start of the run; expensive components should be flagged so the shop handles them carefully rather than treating them as attrition.[24] In one high-volume factory the machine’s dropped parts were collected at the end of each shift and returned to reels, which is usually harmless, but a lead bent in the drop produces a part that no longer mounts flush.[328]
The feeder behaves as a smart cartridge that completes the machine: it carries identifying data about what part it holds, so the machine knows the contents of each slot without separate programming.[411] Powered eight-millimetre and twelve-millimetre tape feeders that identify themselves let the machine scan for all loaded feeders and remember which part is in each, so feeders can be swapped between positions without reprogramming.[686]
Alignment and vision
Two alignment technologies are used: a camera and a laser. Components sit loose in the tape pocket and are never picked exactly centred, so the machine must measure what it actually holds; one method rotates the part ninety degrees past a laser line to derive both the angular and the XY offset and corrects for them at placement.[153] A through-axis camera that looks along the nozzle is the feature that separates a working machine from a toy, because without it there is no way to see what the vacuum pickup is holding in order to align it.[153]
Cheap machines substitute stepper motors for the closed-loop positioning that placement accuracy requires; good encoders are what make a machine accurate and they are expensive, so a capable machine below five thousand dollars remains out of reach.[299] Most placement machines work natively in metric and convert Imperial input, so a design submitted in inches accumulates cascading rounding errors through the conversion.[299]
Board referencing varies. Some low-cost machines reference the board from the first component’s position rather than from fiducials, which changes the setup procedure compared with fiducial-based alignment.[477] Fiducial support is a differentiator among low-cost machines: one sub-three-thousand-dollar model supports only a single fiducial, which limits how far board scale and rotation can be corrected.[610] On a desktop machine, the single most important calibration number is the XY offset in the vision plane from the centre of the top camera’s field of view to the point where the nozzle tip touches, because fiducial calibration locates the board in camera space while the nozzle must be commanded in real-world space; there are six or seven such key numbers and each machine differs by around two hundred microns.[686]
Placement process and failure modes
Reliable picking is tuned per component: nozzle dwell time on the exposed tape pocket, suction level, and release timing are all separately adjustable, and each has to be set for each part type.[419] If the Z height is set wrong the head presses harder than intended and parts are flung off the board rather than placed; height and placement pressure are the same adjustment in practice.[428] Better machines allow the placement force to be programmed, which matters because delicate parts, such as some capacitors, have their leads damaged if the head sets them down too hard.[188]
On a run of roughly twelve hundred identical LEDs, a desktop machine placed about ninety-five percent correctly with little setup; the residual failure was parts landing on their side, which the vision system did not reject because the lens made the part outline non-square and it failed to recognise the wrong-side silhouette.[419] On a desktop machine running a batch of five hundred boards, an error that halts the run occurs on roughly one board in ten, or about one placement in a few hundred; the usual causes are the vacuum head sitting on the part badly or the upward-facing camera finding a silhouette that does not match the expected part size or orientation.[716] A machine can be set to abandon a part after several consecutive vision failures, complete everything else on the board, and report the omissions at the end, so a single bad feeder does not stop the run ten minutes in and waste the remaining hours.[224]
Roughly half the labour of populating boards is loading the correct components into the correct feeders, and a mix-up such as a ten-kilohm resistor in the slot programmed for a one-hundred-nanofarad capacitor is catastrophic for the whole run.[157]
Component packages
Every placement machine handles 0603 as a baseline, but 0402 is a discontinuity: it requires different, smaller nozzles, not all machines can do it, and yield falls, so a design should not go below 0603 without a reason.[104] Going below 0402 to 0201 requires smaller nozzle heads again, which not every assembler owns; a shop that must buy nozzles for one customer’s job will charge the tooling back as non-recurring engineering.[96]
Package choice is a filter on how many suppliers can build the board: at 0805 any machine will do the work, 0402 cuts the available machines to roughly sixty-five percent, and 0201 to roughly thirty percent.[502] A machine specified for 0201 may not achieve it in daily production, because motor repeatability at that precision is not the same thing as placing real parts from real feeders every day, and a machine that has only ever been run on larger passives by its existing customers has not been proved at the small end.[237] Even where a machine can place 0201, the precision mode can add on the order of eighty percent to the time per board, which does not show up in the placement specification.[237] Choosing the smallest possible packages to minimise board area constrains manufacturing at both ends: it narrows which machines can be bought and it rules out many contract manufacturers.[237]
The influence runs in both directions: components intended for automated assembly are supplied in tape and reel because that is the format compatible with placement machinery, which is why single-board computer modules are sold on wide reels for machine placement.[563]
Board and panel design
A panel must be sized to the assembler’s machine, not to the PCB fabricator’s maximum panel; fabricators quote panels in inches, commonly eighteen by twenty-four, and panels of that size frequently do not fit a placement machine.[415] Board width, not length, is usually the binding limit, because boards feed lengthwise through the machine on rails; one desktop machine offers about four hundred millimetres of total span and its rails cannot be set to the extreme edges.[415] A large panel held only at its edge rails will bow when the head presses down, and a V-grooved or slot-routed panel of that size can fold or snap in half at the first placement.[415] Panelisation should therefore be agreed with the assembler before the boards are fabricated, since a panel that does not fit the placement machine is discovered only when the finished panels arrive at the assembly house.[149]
The economic unit a placement machine works on is a panel rather than a single board, so a machine whose build area is only Euro-card sized defeats the purpose of automation.[148] Owning a placement machine changes layout practice permanently in this respect: every board is panelised from then on, because panelising costs nothing and a machine run on a panel is proportionally cheaper to set up.[403] For one five-hundred-millimetre circular LED display, Mike Harrison built the board as four separate quadrant boards of about two hundred and fifty millimetres square specifically so each piece would fit his placement machine.[224] Large-format assembly machines exist that take boards up to about one point six metres in length; they are built as a widened standard machine with two heads that can either run as two independent machines on small boards or both place onto one long board, and they are expensive per board for small work.[224]
Programming and setup
The centroid is the reference point the machine picks a part at, so a library footprint whose origin sits on pin one forces the assembler to re-enter a reference point for every such component; the pin-one marker is of no use to the placement program.[408] Placing the origin at the part centroid conflicts with layout practice, because the designer works on a placement snap grid and a centred origin leaves the pins off the routing grid.[408] A component type the machine’s library has not seen before still requires a person to read the data sheet and enter the part dimensions; this step of the flow has not been automated.[411] Setting up a machine for a new board requires a pilot production run to confirm that the program, the part orientations, and the rest of the configuration are correct before committing to the batch.[257]
Workflow choices reduce this burden. Harrison keeps all of his PCB library orientations identical to the tape orientation in his feeders, so a previously used part never has to be checked for rotation when it is placed again.[412] His short-timescale workflow runs layout first, exports the bill of materials from the board file as the shopping list for parts, boards and stencil, and begins assembly on the in-house machine about two days later.[224] An in-house computer-aided-manufacturing workflow can analyse the Gerber and pick-and-place files and generate the machine programs directly, with release management and file tracking held in the same system as the rest of the manufacturing data.[722]
Throughput and line balance
Hand placement runs at roughly three hundred components an hour, which is the figure a desktop machine has to beat; a machine placing one hundred components an hour is useless.[317][224] One desktop design claimed twelve hundred and fifty components an hour.[317] On an eight-by-eight panel carrying two hundred and fifty-six parts, Ryan O’Hara found that hand placement took about half an hour while a single-head machine completed it in under three minutes.[153] A run of twelve hundred identical LEDs across a four-up panel took roughly five to ten minutes per board with three placement heads working, about twenty minutes per panel.[428]
Setup dominates infrequent use: exporting the position file, converting it to the machine’s format and recalling the procedure took two to three hours for one job, principally because the operator does not do it often enough to retain the steps.[428] For a small job the machine setup can exceed the time to place the board by hand, and the arithmetic should be done before the machine is set up rather than after.[477]
On a mixed-job line the bottleneck is set by process mix rather than machine speed: at Macrofab, leaded chip parts slow placement while they are validated, bulk passives run fast, and on some jobs reflow is the slowest step.[243] Where a job uses the assembler’s house-stocked parts, only the few remaining components are hand-placed, at roughly ten seconds per part.[243] With a semi-automatic stencil printer the board is loaded by hand and the stencil swap is the slow step, so printer setup can take longer than the placement machine setup for the same job.[243]
Line productivity is decided by the automation around the machines rather than by the machines themselves: buffers, magazines and loaders keep the machine fed and unloaded so nobody has to stand at it, on the argument that machines are cheap relative to people.[411] A rolling changeover keeps a line running: the upstream stencil printer is changed over and the next job printed while the placement machines are still finishing the current one, and each placement machine is changed over while the one after it still runs.[411] A board with around eight hundred and fifty placements drawn from about a hundred and five unique parts cannot be built in one pass on a single machine, so two or three placement machines are run in series after a single paste print.[411] The stated ideal for a contract assembler is that every step from stencil printing through placement and reflow happens without operator intervention, so labour is confined to what arrives and what ships.[411] Even at professional assembly houses operators are commonly seen constantly attending the machines; where the process is correctly set up they should not be, and a well-set-up shop can have one person running three or four machines, which requires either high volume or a predictable repeating schedule with a dedicated bank of feeders per customer.[224]
Economics of ownership
Placement equipment fell from roughly half a million dollars to around ten thousand dollars over the period in which small-scale assembly became feasible, which put the equipment within reach of small companies and individuals.[6] High-end multi-head production machines cost on the order of half a million dollars, and capital equipment for an assembly line at a million dollars a machine is not unusual, with a sales and evaluation cycle running close to a year.[411] A bare-bones new desktop machine can be had for around ten thousand dollars, but feeders and the surrounding support equipment take the real figure to twenty thousand and upwards; the machines used by the better-known open-hardware companies were around fifty thousand.[126] Running a machine of that class is a full-time job in itself, so a single-person operation gains nothing from owning one and loses the automation it was bought for.[126]
The purchase decision is an arithmetic comparison against per-board assembly pricing: at roughly five dollars a board to have it assembled outside, a ten-thousand-dollar machine plus the time to learn, run and maintain it needs a large and steady volume to be recovered.[612] Because programming, reeling and loading are fixed overheads that only pay back over many identical boards, a placement machine is justified by commercial production quantities rather than by prototyping.[273] A rough quantity threshold for setting up a machine is around one hundred boards; at ten boards the setup is not recovered and hand placement is faster.[195] Above roughly one hundred units the case for in-house placement weakens and outside assembly becomes the sensible route; below it, an owner with a heavily optimised flow and repeat clients can beat the turnaround an outside shop will give.[319] A workshop making batches of about fifty hand-assembles instead, because the barrier is not only the purchase price but the time to learn the machine and the floor space it needs; outsourcing at five to ten pounds a board is cheaper than absorbing both.[643] A machine that is not used at least weekly is a poor allocation of capital compared with other equipment, whatever else it teaches its owner about manufacturing.[270] The viability of a low-end machine is decided by placements per hour against manual placement, the number and reliability of the feeders, the setup time, and the machine’s reliability; the combination leaves only a narrow window in which owning one saves money.[63] Of the people who believe they need a placement machine, roughly ninety percent do not; the genuine case is low board counts, high product mix and a requirement for fast turnaround, conditional on accepting the time needed to develop the process and software flow.[412]
The subcontractor problem for short runs is structural: very small assembly shops fail when they lose a single customer, large ones will not take small jobs, and a correctly sized one will schedule small jobs around its major customers, with no option to pay a premium for a faster turnaround because the job is not worth their while.[135] Owning the machine also removes the documentation burden that a subcontractor requires, which for short-timescale one-off work can dominate the job.[224] A machine may sit idle for a month between jobs and still be worth owning where the alternative is subcontractor scheduling.[412] Automation does not remove labour from assembly so much as change it: loading a placement machine requires organisation and a stockroom, and the facilities that house such machines are of a higher standard than hand-stuffing operations.[143] The difficulty of running a placement machine is not one large problem but a large number of very small ones, each individually capable of stopping a job, and they are only learned by using a machine.[412]
Ownership feeds back into design practice. Bringing placement in house converts an electrical engineer into a process engineer, whose concerns become solder-paste thickness, placement accuracy to the thousandth of an inch, rotation, and the statistical distribution of placement over time.[232] The designer learns to avoid parts that would require a nozzle change and steers component selection by what the machine handles well.[153] For quantities of two or three pieces the shortest lead time is to build the boards in-house with a tabletop stencil printer and reflow oven, while contract shops that maintain a prototype line remain the route for anything larger.[232]
Several documented operations illustrate the arithmetic. Vincent Himpe’s operation bought a used Juki on eBay for forty thousand dollars, with eighty tape-and-reel feed lines, stick feeders and a tray feeder, and paid for it within four months against what the same volume of boards would have cost in subcontractor charges and labour; the surrounding process was a small reflow oven, a manual screen printer, a full stock of reels and a dedicated operator.[169] Chris Anderson’s operation found that outsourcing at a batch size chosen for unit economics froze the design until the inventory was consumed, while an in-house line making about a hundred units a day removed that constraint and allowed roughly thirty board revisions in a year, each one cheaper or more reliable to build.[105] Charles Alexanian’s company, which makes everything in batches of fifty to a hundred, justifies its machine because all of its products fit on one set of feeders, so the setup can be left standing between jobs; offering the same machine as a service to outside customers would destroy that advantage.[429] The machine works but needs continual attention, and each time it is restarted after a gap the operator relearns another of its quirks.[429] One threshold used in practice at Seeed is a hundred to five hundred units built in-house on the shop’s own machine for speed, with anything above a thousand sent to specialist assembly houses that do nothing else and achieve higher yields.[125]
Equipment market and classes
Used mid-1990s production machines running DOS remain capable and can be bought in good condition for around ten thousand dollars, with about fifteen thousand covering the machine set up and supported.[153] A refurbishment market exists for older production machines: a vendor takes them in, replaces ageing components such as hard drives with more stable parts, and resells them, which is how a forty-eight-feeder machine reaches a small shop.[270] Used desktop machines are scarce outside the main markets: in Australia they come through one or two large dealers at around twenty thousand dollars, and a used machine under five thousand is not realistic.[612] A nominally five-thousand-dollar imported machine landed in Australia at about six thousand once postage and currency conversion were counted, the same money as a large quantity of assembly time at a professional shop.[178] A four-thousand-dollar Chinese desktop machine reaches about five to five and a half thousand delivered once shipping and a spare-parts kit are included, and at that point it lacks the vision and alignment capability that would make it useful.[153] A genuinely capable machine below five thousand dollars is not achievable, because a machine at that price will not be large enough or structurally rigid enough to carry ten-thousand-part reels.[299] Akiba of Freaklabs bought a twenty-five-thousand-dollar Japanese desktop machine, made by MDC in Tokyo, as a used floor model; it was designed for prototyping rather than manufacturing, where production-grade placement means Juki, Panasonic or Samsung equipment.[245]
Newer low-cost machines have narrowed the gap. A Neoden 4 desktop machine carries forty-eight feeders with both up-looking and down-looking vision and places 0402 components, at around ten thousand dollars.[372] A later model brought a dual-camera, dual-head machine with an automatic tool changer and forty feeders under three thousand dollars, but its feeders are built in rather than removable cassettes, so a feeder cannot be preloaded off the machine or swapped between jobs.[610] David Ray’s operation runs three low-cost Charmhigh machines in series to cover a bill of materials larger than one machine’s feeder capacity, or runs the same job on all three in parallel; most jobs have fewer than fifty bill-of-materials lines, and of those a further portion is through-hole or too large for the machine to place.[716] Assist machines occupy the space between hand and automatic placement: an XY table with a thumb-pressed vacuum head, around sixteen feeders and tube support, moved by the operator’s arm, at roughly four thousand dollars.[508] A cheap machine can only take tape feeders, so a board with parts in tubes or trays cannot be fully populated on it, and if the machine cannot place every package on the board the case for using it collapses.[148]
Stephen Hawes designed the Lumen, a desktop open-source machine of this class, for low to medium volume, specifically in-house prototype batches for teams unwilling to wait around ten days for an outside assembler to return boards.[686] Placement-machine software is poor across the industry because switching cost locks a buyer to a fifty-thousand-dollar machine regardless of the software, and vendors compete on mechanical specifications such as transit speed and parts per hour, with a field application engineer supplied to make the software work.[686] Hawes built the machine on an existing 3D-printer firmware stack as a deliberate reuse decision rather than a technical necessity: the motion problem is the same and the host firmware already has years of contributed work behind it.[686]
Import tariffs of twenty-five percent applied to manufacturing equipment as well as components, changing the delivered price of a desktop placement machine by about twenty-five hundred dollars depending on the order date.[400] Floor space and noise are first-order constraints on owning placement equipment: a full line takes a large amount of room and is loud, and machines have been rejected purely because they would not fit through the door.[613] A machine left unused still needs maintenance: running it a couple of times a month to move the motors, and storage covered and out of dust, since the linear drives are the part that degrades.[587]
Assembly services built around the machines
A three-day turnkey assembly service is made possible by constraining the customer to a published parts library: the vendor holds that stock, the parts are already programmed into the feeders, and there is no exposure to outside part delays; Seeed Studios announced such a service on those terms.[255] Turnkey assemblers price lower for designs built from their suggested parts because those reels are already loaded in their machines; the saving is unavailable to designs whose parts are precision or single-source and cannot be substituted.[682] CircuitHub began with the thesis of driving customers onto roughly two thousand common parts and keeping twenty-five to thirty machines permanently loaded with them; the size of the part space defeated it, and the working model instead decouples mass storage of individual parts from the placement step and treats material handling as the central problem.[699]
Other applications
Fan-out wafer-level packaging uses a placement machine in the same role as board assembly: singulated die are placed onto double-sided adhesive tape on a steel carrier with deliberate gaps, then over-moulded in epoxy to form a reconstituted plastic wafer with a flat top surface for subsequent redistribution layers.[469]