中文

Can a mounter pick up a part bigger than a palm and heavier than a phone?

SMT process Q&A · large parts and nozzles

AI servers are sending packages of 150–200 mm and half a kilogram to two kilograms into SMT lines, while the catalog limit of general-purpose mounters stops at 135 mm square and 350 g. This article splits “pick, move, recognize, place” into four gates and works out each one: it looks up the manufacturer data that can be looked up (eight manufacturers, 12 models), and estimates what cannot be found with public engineering formulas.

Jiahon affiliated company, Engineering Department · SMT nozzle technical file

A process engineer handed a package wider than a palm and noticeably heavy in the hand — for example the 175 × 175 mm, 650 g class — usually asks two things first: can the machine take it, and how should the nozzle be made? Looking at the catalog, for a mainstream general-purpose mounter such as the Panasonic NPM, the standard head limit stops at 120 × 90 mm and 50 g, and the option reaches 135 mm square at most and 350 g. The numbers do not match, and the matter is stuck.

The reason for the stall is mostly that the “machine limit” is read as one number. Picking and placing a part is actually four motions, and the physical quantity that holds each gate is different: picking depends on the Z-axis suction force, moving depends on XY-axis friction, recognition depends on camera range and feature points, and placing depends on load. Worked out separately, the answer changes from “no” to “which gate needs what added”.

The four gates of pick and place

Figure 1. The four gates of pick and place. Gates ① and ② are set by the suction force at the nozzle end and can be estimated with formulas; ③ and ④ belong to the machine’s specification and program settings.

Jiahon does not sell mounters. What we make is nozzles, so we can put the public specifications of each maker’s machines in one table and compare them, without speaking for any one machine.

The method is stated first: where a manufacturer has published data, we quote it with its source and viewing date; where it has not (vacuum value, acceleration), we estimate with public engineering formulas and mark the result as an estimate. Of the ten questions below, the first eight follow the order of the four gates, the ninth explains the estimating method, and the tenth is a list of data to prepare.

Questions in this article

  1. How big and how heavy counts as a “large part”? Where do the public limits of each mounter stop?

  2. How large can a nozzle be made?

  3. Picking: how much suction force is needed at the instant the Z-axis lifts?

  4. Moving: after pick-up, will the part be thrown off in high-speed motion?

  5. How much vacuum? What percent of the part’s top surface must the suction area take?

  6. Recognition: the part is bigger than the camera’s field of view; can the machine recognize it?

  7. Tens of thousands of solder balls: can the machine count them all?

  8. Placing: how much force to press?

  9. What if the manufacturer has not published the data? Parallel estimation from public sources

  10. If you want a custom nozzle for a large part, what data should you prepare?

Q 01. How big and how heavy counts as a “large part”? Where do the public limits of each mounter stop?

Taking the Panasonic NPM series as the example, the standard limit of mainstream general-purpose machines is 120 × 90 mm and 50 g, with an option up to 135 mm square and 350 g; for models or heads launched for AI servers, the public values have reached 200 × 200 mm and 2 kg.

In this article a “large part” means a package beyond the standard head limit of a mainstream general-purpose mounter — the class with a side longer than 120 mm and a weight over 50 g; large BGAs, substrate-level packages and large connectors on AI servers all fall here. Table 1 puts the public limits of eight manufacturers into one set of columns; where nothing is published it says “not published”, with no filling in and no guessing.

Component size limits published by each manufacturer, to the same scale and sharing the lower-left corner

Figure 2. Component size limits published by each manufacturer, to the same scale and sharing the lower-left corner. The dashed frame is the Panasonic NPM option limit of 135 mm square; the solid frames from inside out are Panasonic standard, JUKI, ASMPT and FUJI.

Table 1. Large-part limits published by each manufacturer (one set of columns; “not published” is written as it is)

Manufacturer / model, headSize limit (mm)Height (mm)Weight limitPlacement forceLarge-part recognitionData version
Panasonic NPM-W2, 3-head V2120 × 90 or 150 × 25; option 135 sq.T30 (option T40); T13 at 135 sq.30 g; option 300 g100 NSplit recognition not listed in the catalogCatalog 2024-05-31
Panasonic NPM-WX, 3-head V2 / 4-head120 × 90 or 150 × 25T4050 g100 NSplit recognition not listed in the catalogCatalog 2022-04-25
Panasonic NPM-GW, 3-head120 × 90 or 150 × 25; option 135 sq.T40; T18 at 135 sq.50 g; option 350 g100 NSplit recognition not listed in the catalogCatalog 2025-01-16
ASMPT SIPLACE SX2 TWIN, VHF head175 × 175 (240 × 240 coming soon)T50650 g (2.0 kg coming soon)0.5–100 NCamera field of view 66 × 50 mm, can recognize up to 200 × 200 mm; can inspect 50,000+ solder ballsTechnical document 2026-05
FUJI NXTR / AIMEX, standard configuration (head not specified)200 × 200Not published2.0 kg (demonstrated under evaluation conditions; 1.3 kg achieved)Not listed in the news releaseSplit recognition of large partsNews release 2026-07-30
FUJI NXTR S / A, AIMEX, RH01 head200 × 150T38.1Not published for each head (DX head: 250 g track record)Head not listed; the same maker’s NXT III / AIMEX III press-in placement 3.5–100 NSame as aboveWebsite 2026-09
Yamaha YRM20DL, LM head90 × 139T40500 gForce control 100 NThe same maker’s YSM series such as the YSM20R: split recognition for W45 mm and aboveWebsite 2026-09
JUKI JM-E01, 2-axis insertion head150 sq.T90400 gInsertion load 50 N3D image splitting: 40 × 25 per shot, 4 × 3 reaches 150 × 71Website 2026-09
JUKI RS-2, with 54 mm field-of-view camera50 × 150 (1 × 3 split); 74 sq. (2 × 2 split)Not publishedNot publishedNot publishedSplit recognition (1 × 3, 2 × 2)Catalog 2024-12
Universal Fuzion, OF / Z7 / FZ4 head150 sq.T40Not published49 N (5 kgf)Not listed in the catalogDatasheet MC-7490
Hanwha XM520, 5-axis × 2 gantry150 × 75T55300 gNot publishedSupports split recognitionWebsite 2026-09
Mycronic MYPro A41, MIDAS head99 × 73T15140 gNot publishedLine-scan camera field of view 80 mmDatasheet 2025-01

Units in mm; “sq.” means the side of a square, T is part height. All values are public manufacturer data (viewed 2026-09-19), obtained under each manufacturer’s own mechanism and control conditions; one maker’s numbers cannot be used to infer whether another maker’s machine can do it. The shaded rows on the Chinese page are the sources of this article’s two worked examples. This is the general case, and individual cases still vary.

Three things can be read straight from this table:

  • Size: the highest public values are FUJI’s 200 × 200 mm (2026-07 news release) and ASMPT’s 175 × 175 mm; the JUKI JM-E01 is 150 mm square; Panasonic’s three mainstream NPM machines reach 135 mm square with the option.

  • Weight: the highest public values are FUJI’s 2.0 kg (demonstrated under evaluation conditions), ASMPT’s 650 g and the Yamaha LM head’s 500 g; most manufacturers do not publish weight, and none of JUKI’s mounter series states it (the insertion machine JM-E01 publishes 400 g).

  • What nobody publishes: machine acceleration is published by Universal Instruments alone (platform of the 2.5 G class), and no one publishes the vacuum value — these two are exactly the numbers needed to work out suction force, and Q09 explains how to fill them in.

Three manufacturers say why this round of specifications was raised: FUJI’s news release and ASMPT’s AI Server Applications technical document state outright that it is for large, heavy packages on AI servers, and Yamaha’s LM head description cites high-current EV connectors and the large BGAs brought by generative AI. In other words, this is not a special need of one customer, but a specification band that the whole industry is moving into.

Q 02. How large can a nozzle be made?

The limit is not in the nozzle but in the head — the pitch P between adjacent nozzles. If the nozzle outline exceeds P, it collides with the neighbouring nozzle.

How large a custom nozzle’s outline may be is, in theory, limited by one geometric condition: the nozzles on a head are arranged at a fixed pitch P, and once a nozzle outline exceeds P it interferes with the neighbouring nozzle or neighbouring part during pick or place. So the answer to “how large can the nozzle be” is “within P”.

Upper limit of nozzle outline = pitch P between adjacent nozzles

Figure 3. Upper limit of nozzle outline = pitch P between adjacent nozzles (schematic). P depends on head type: 16-head, 12-head, 8-head, 4-head and 3-head each differ.

P is set by the head type. High-speed heads with many nozzles have a small P; head types that handle large parts have few nozzles and a large pitch — the Panasonic NPM 3-head V2 / 4-head and Yamaha’s LM head (3 nozzles) are of this kind, and FUJI’s RH01 is a single-nozzle head with no neighbouring nozzle problem, so the limit returns to the mechanism itself. So the first step in assessing a large-part nozzle is not to draw the nozzle, but to confirm which head type this part will go on and get P in hand. Jiahon’s Panasonic-fit 3Head Special nozzle series is made for Panasonic’s 3-head series, with the outline set to that head type’s P; for items for general models see the Panasonic mounter nozzle overview.

How large the nozzle end face itself must be is a different question, set by suction force, and the next three questions work that out. If your part is the small and tall kind with no place to pick on the top surface, go straight to The part is taller than the board: where should the nozzle pick?, which covers the lower and upper limits of end face size.

› Related products: SMT nozzle overview

Q 03. Picking: how much suction force is needed at the instant the Z-axis lifts?

At the instant of pick, just two things matter: the part’s weight and the Z-axis upward acceleration. A 650 g part needs about 32–64 N at 1–3 G.

The nozzle descends, touches the part, builds vacuum, and then the Z-axis lifts the part. The suction force at this gate has to resist gravity plus the inertial force of the upward acceleration, multiplied by a safety factor:

F_required = m × (g + a) × S

m is the part mass, g gravity 9.81 m/s², a the Z-axis acceleration, S the safety factor (this article takes 2.5: Schmalz asks for 2.0 or more for critical, rough or oily workpieces; the factor of 4 in SMC’s selection guide is a static calculation that folds acceleration in, and this article lists acceleration separately, so it is not counted twice).

Putting in the two limits with published weights from the Q01 table, with Z acceleration of 1–3 G:

Table 2. Suction force and suction area needed for picking (Z-axis)

Worked exampleAccelerationSuction force neededSuction area needed (50 kPa)Share of part top surface
175 × 175 mm / 650 g (published limit of ASMPT SIPLACE SX2 TWIN VHF)1 G31.9 N638 mm²2.1%
2 G47.8 N956 mm²3.1%
3 G63.8 N1,275 mm²4.2%
200 × 200 mm / 2.0 kg (value published in FUJI’s 2026-07-30 news release)1 G98.1 N1,962 mm²4.9%
2 G147.2 N2,943 mm²7.4%
3 G196.2 N3,924 mm²9.8%

Values put in: g 9.81 m/s², safety factor S 2.5, usable vacuum at the nozzle end 50 kPa; weights are each manufacturer’s published limit. This table is an engineering estimate, not a promised value for any machine or nozzle; this is the general case, and individual cases still vary.

The demand at the picking gate is actually not high: even lifted at 3 G, a 650 g part needs just 64 N, which as suction area is 4.2% of the top surface. What is really tight is the next gate.

Q 04. Moving: after pick-up, will the part be thrown off in high-speed motion?

While moving, the part follows the nozzle by friction, and at the same G the suction force needed is higher than for picking — a 650 g part needs 48–112 N at 1–3 G, and that is the value nozzle design should take.

Forces in picking

Figure 4. Forces in picking (left) and moving (right). When moving, suction force does not directly resist the inertial force; it first has gravity taken off and is then multiplied by the friction coefficient, and just what is left is the force that can carry the part along.

After the part is held, the head accelerates and decelerates in the XY direction. Now vacuum just presses the part against the nozzle pad, and what really keeps the part from sliding is the friction between pad and part. Following Schmalz’s theoretical holding force of a suction cup (load case II, horizontal transport):

F_required = m × (g + a ÷ μ) × S

μ is the friction coefficient of the nozzle pad against the part’s top surface; this article takes 0.5 (dry, clean plastic or metal surface); oil or an aged silicone pad would be lower.

The term a divided by μ is why moving is harder than picking: with μ at 0.5, every 1 G of horizontal acceleration has to be paid for with 2 G of suction.

Table 3. Suction force, suction area and equivalent round nozzle needed for moving (XY axes)

Worked exampleAccelerationSuction force neededSuction area needed (50 kPa)Share of top surfaceEquivalent round suction face
175 × 175 mm / 650 g (published limit of ASMPT SIPLACE SX2 TWIN VHF)1 G47.8 N956 mm²3.1%Φ35 mm
2 G79.7 N1,594 mm²5.2%Φ45 mm
3 G111.6 N2,232 mm²7.3%Φ53 mm
200 × 200 mm / 2.0 kg (value published in FUJI’s 2026-07-30 news release)1 G147.2 N2,943 mm²7.4%Φ61 mm
2 G245.3 N4,905 mm²12.3%Φ79 mm
3 G343.4 N6,867 mm²17.2%Φ94 mm

Values put in: g 9.81 m/s², safety factor S 2.5, usable vacuum at the nozzle end 50 kPa; weights are each manufacturer’s published limit. Friction coefficient μ 0.5 (nozzle pad against the part’s top surface, Schmalz’s recommended value). This table is an engineering estimate, not a promised value for any machine or nozzle; this is the general case, and individual cases still vary.

Comparing the two examples shows that “heavy” is harder than “big”:

  • A 650 g part at 2 G needs an equivalent Φ45 mm suction face, 5.2% of the top surface — the top surface of a 175 mm square can hold it, as long as there is a clean flat area in the middle.

  • A 2.0 kg part needs an equivalent Φ79 mm at 2 G and Φ94 mm at 3 G — this is already close to the order of P for 4-head and 3-head types, and the top surface of the part must have an unobstructed flat area that large. That is why, for placement at the 2 kg class, manufacturers use special head types and slow-down modes, not an ordinary nozzle sucking hard.

Trading area for acceleration

The suction area the nozzle end can supply is limited (the upper limit is P, and the obstructions on the part’s top surface also limit it), while the machine’s acceleration can be adjusted in the program. Large parts are mostly placed in the machine’s low-speed or large-part mode: first fix the cycle time the line accepts, then work back to how large the nozzle must be — the order cannot be reversed. The gap between the CPH in the catalog and real cycle time is worked out in The speed on the catalog is not your speed.

Q 05. How much vacuum? What percent of the part’s top surface must the suction area take?

With a stable air supply, a vacuum generator gives a stable vacuum value; taking a conservative 50 kPa, a 650 g part at 2 G needs about 1,594 mm², 5.2% of the top surface; 2 kg at 2 G needs 12.3%.

Suction force equals vacuum times the actual suction area: F = ΔP × A. A mounter’s vacuum comes from a vacuum generator (ejector), and as long as compressed air is supplied steadily, the vacuum value is steady — the air supply condition published in the catalogs of the three Panasonic NPM models is 0.5 MPa or more at 200 L/min; the upper limit in a commercial vacuum generator catalog is −84 kPa (SMC ZM series standard model, supply 0.5 MPa), and the same catalog notes that the vacuum value falls as leakage increases.

This article does not use the catalog’s upper limit, but conservatively takes 50 kPa as the usable vacuum at the nozzle end, because between the generator and the nozzle there are still the pressure losses of filters, rotary joints and long tubing, and leakage between pad and part, which can never seal perfectly. This value leaves a margin of about forty percent.

The equivalent round suction face needed for moving

Figure 5. The equivalent round suction face needed for moving (XY axes), to the same scale as the part’s top surface. Left: 175 × 175 mm / 650 g; right: 200 × 200 mm / 2.0 kg. The three rings from inside out are 1, 2 and 3 G.

Why the rough-screening formula for small parts cannot be reused

Selection of small parts often uses a rough screening formula: weight × 10 G × a margin of 5 ÷ 20 kPa (the article The part is taller than the board uses exactly this one). It pushes acceleration, safety factor and leakage all to the extreme, and works well for parts like 0201 and 2012, whose top surface is hardly restricted — the sealing area it gives is under 1 mm², and even being this conservative causes no trouble.

But the same formula applied to a 650 g part asks for a suction face of 15,941 mm², 52% of the top surface. It is not that the machine cannot do it; the rough formula has stopped working here: the top surface of a large part is a scarce resource (there are parts, heat sinks, labels and warpage), and acceleration is something the machine lets you set, so the method of Q03 and Q04, which works out each load case separately and treats G as a variable rather than a constant, must be used instead. The two articles differ in range of use, not in physics.

Two values that must be measured

The actual vacuum value at the nozzle end and the friction coefficient of the pad count just when measured on your machine. This article’s 50 kPa and μ 0.5 are assumptions for estimating; when designing a nozzle, Jiahon will ask you for the reading of the machine’s vacuum gauge, or replace the assumptions with a trial pick on the actual part.

Q 06. Recognition: the part is bigger than the camera’s field of view; can the machine recognize it?

Yes, provided a “unique” feature point can be found within the camera’s range. The whole part coming into view is not a necessary condition — ASMPT publishes a camera field of view of just 66 × 50 mm, yet it can recognize a 200 × 200 mm part.

After a part is picked up, it first passes the machine’s recognition camera, which measures the part’s offset and angle against the nozzle center, and the machine corrects the placement coordinates accordingly. A large part is often bigger than the camera’s single field of view, and the approach of each maker is split recognition: take several images and stitch them, or take just a few local areas of the part to locate it. Yamaha publishes split recognition for W45 mm and above; JUKI publishes 1 × 3 and 2 × 2 splits and a 4 × 3 split with 3D images; FUJI publishes split imaging of large parts; Hanwha publishes support for split recognition; ASMPT publishes a part camera field of view of 66 × 50 mm and recognition of parts up to 200 × 200 mm.

For Panasonic, the range of part data that can be built for the Offline Camera Unit (which uses the same multi-function recognition camera as the current machines) is published as 0201 to 135 mm square, height 65 mm; how split recognition works is not described in the catalog or the website specification page (viewed 2026-09-19). Please confirm this item directly with the machine manufacturer.

The feature point must be unique

 Left: just regularly arranged solder balls — even if the whole pattern shifts by a row, every ball still matches, and the machine may place

Figure 6. Left: just regularly arranged solder balls — even if the whole pattern shifts by a row, every ball still matches, and the machine may place the whole part off by one pitch. Right: a unique feature point within the camera’s range (for example, a missing-ball position) — after a shift, four places no longer match, so the position is unique.

Split recognition solves just the problem of “cannot see all of it”; what decides whether it can be located is whether the camera’s range holds a unique feature. A BGA’s ball array is regular: if the field of view holds just a neat patch of balls, shift the whole array by one row and every ball still matches a ball, and the machine cannot tell — this is an old machine-vision problem: a Cognex patent lists “cannot be judged as unique” as a case that must report an error, and a KLA patent requires the alignment feature to be unique in both the X and Y directions. The result is a whole part placed off by one pitch, which SPI / AOI catch only later.

So before going on the machine, do one very cheap thing: open the bottom view (ball map) in the part datasheet, or simply take a photo of the underside of the part, and see whether the camera’s range holds any of these: missing-ball positions, ball-free zones, decoupling capacitors on the underside, screen-printed or laser marks, asymmetric corners. If it does, recognition has something to locate by; if not, discuss with the machine manufacturer switching to outline recognition or adding a fixture. Whether a feature can be adopted by your machine’s recognition program depends on the manufacturer’s reply and verification on the real machine.

The cost of being off by one pitch

For a BGA with 0.8–1.0 mm pitch shifted by a row, the whole line of balls lands on the neighbouring pads; this defect is almost invisible before reflow, and what gets reworked is a whole high-value substrate. Looking at the ball map first is one of the cheapest of the four gates.

Q 07. Tens of thousands of solder balls: can the machine count them all?

A manufacturer publishes that it can inspect 50,000 or more. The count itself is not the limit; pitch is: putting 50,000 balls into a 200 mm square gives a pitch of about 0.90 mm, within the manufacturer’s published range of 0.8–1.0 mm.

Most manufacturers’ catalogs do not state an upper limit on ball count. The public value found so far is ASMPT’s statement in its AI Server Applications technical document of “reliably inspects 50,000 or more solder balls”, and the same document says that these HPC BGAs have ball diameters of 0.5–0.6 mm and pitch of 0.8–1.0 mm.

The count can be turned back into pitch: 50,000 balls evenly in a square array is about √50,000 ≈ 224 balls per side; within a 200 mm square, pitch ≈ 200 ÷ 223 ≈ 0.90 mm — consistent with the pitch band the manufacturer publishes, and it also means 50,000 is the natural result of a “200 mm square, 0.9 mm pitch” package, not an especially high threshold.

What really affects recognition is three things: whether the pitch is within the camera’s resolution (ASMPT publishes lower limits of 140 / 300 μm for ball pitch and 80 μm for ball diameter, and 0.9 mm is far within range), whether the reflection of the ball surfaces is uniform (oxidation and flux residue can make some balls vanish from the image), and how much cycle time the number of split images takes. The third goes back to the conclusion of Q04: placing large parts already runs in low-speed mode, so imaging time is usually not the bottleneck.

Q 08. Placing: how much force to press?

At this gate manufacturers are fairly open: the large-part heads of Panasonic’s three mainstream machines all have a placement load of 100 N; ASMPT 0.5–100 N, Yamaha LM head 100 N, FUJI press-in 3.5–100 N, JUKI JM-E01 50 N.

Placing is the motion of pressing the part into solder paste or into a hole position; the machine sets the load with force control, the program decides it per part, and the nozzle end adds no condition of its own. A large BGA usually needs just a few to a few tens of newtons to make the balls touch the paste firmly; the 100 N class is made for press-fit connectors and large warped substrates.

What the nozzle end must answer is a different question: can the nozzle body and pad bear this load? 100 N is equivalent to 10 kgf pressing on the nozzle, and the body wall thickness, pad material hardness and bonding method of a custom large nozzle must all be designed for this load, otherwise the pad is permanently dented, the vacuum hole deforms, or the nozzle shaft wobbles. That is also why, for items such as the Panasonic-fit 3Head Special nozzle series, Jiahon takes the placement load as one of the design inputs.

With the four gates done, the feasibility of a large part can be summed up in one sentence: picking and moving are set by suction area and G value and can be calculated; recognition is set by feature points, so look at the ball map first; placing is set by load, which manufacturers publish. Of the four gates, what really needs Jiahon and you to work on together are the nozzle design for the first two, and the structural strength of the nozzle at the fourth.

Q 09. What if the manufacturer has not published the data? Parallel estimation from public sources

Not published does not mean it cannot be done. The method has three steps: line up the manufacturers that have published and confirm that physically it is the same matter; use public formulas to work back to the missing values; finally mark the estimate as an estimate and leave it to verification on the real machine.

In putting this summary together, the situation often met was not data that contradict each other, but no data at all: Panasonic does not publish split recognition, vacuum value or acceleration, and JUKI’s mounter series does not publish weight. If you look at one manufacturer’s catalog, the answer is left as “the catalog does not say”. Jiahon handles such questions this way:

Step 1, parallel comparison: put the public values of eight manufacturers into one set of columns (Table 1). Pick and place on mounters is the same physics — vacuum generator, linear motor, area-array or line-scan camera — and a specification that another maker has publicly shown it can do under the same physical conditions is the reasonable capability band for machines of this class. ASMPT publishing 650 g and FUJI publishing 2.0 kg means the 500 g to 1 kg class is technically sound.

Step 2, work back with formulas: where acceleration is not published, use the 2.5 G class, the one public value from Universal, as the upper bound and 1–3 G as the range; where the vacuum value is not published, discount the −84 kPa of the vacuum generator catalog to 50 kPa; put them into the holding force formulas published by Schmalz, and the suction area and nozzle diameter result (Tables 2 and 3). Every value put in is written out, and the reader can swap in the measured values of his own machine and recalculate.

Step 3, mark it as an estimate: an estimate answers just “is there a chance, and at which gate is the gap”, not “your machine can definitely do it”. The last mile of feasibility is the machine manufacturer’s reply and verification on the real machine.

This is also why Jiahon can do this work: we do not sell or act as agent for any mounter, so when we put each maker’s public specifications side by side, we have no stance of defending any machine; what we supply is nozzles and consumables, so there is one thing we care about — to what extent the nozzle end must be made for this part on your machine.

An estimate is not an assurance

All calculated values in this article are engineering estimates, obtained under the assumptions listed in the text; the copyright of each manufacturer’s public data belongs to that manufacturer, and this article cites it as a source. Before adoption, please recalculate with values measured on the real machine.

Q 10. If you want a custom nozzle for a large part, what data should you prepare?

Seven items: model and head type, pitch P between adjacent nozzles, part size and weight, bottom view or photo of the underside, the pickable area on the top surface, the speed the line accepts, and the machine’s vacuum gauge reading (optional).

Every variable in the previous nine questions ends up on the nozzle drawing. The table below collects the design inputs into one list, with the gate where each is used:

Table 4. Design inputs for a custom large-part nozzle

ItemHow to get itGate where it is usedProvided by
Model and head typee.g. NPM-WX 3-head V2, NXTR RH01Decides the pitch P between adjacent nozzles (Q02) and the placement load (Q08)You
Pitch P between adjacent nozzlesHead specification, or measure the nozzle in useGeometric upper limit of nozzle outlineYou / machine manufacturer
Part size, height, weightDatasheet or actual weighingInputs to the suction force and suction area calculation (Q03–Q05)You
Bottom view (ball map) or photo of the undersideBottom view in the part datasheet, or a direct phone photoJudging whether the camera’s range holds a unique feature point (Q06)You
Pickable area on the top surfacePhoto of the top, with heat sinks, capacitors, labels and warpage markedDecides where the suction face can go and whether a porous or odd-shaped pad is neededYou
Cycle time or machine speed mode the line acceptsProcess conditionsDecides the G value taken (Q04): trading area for accelerationYou
Machine vacuum gauge reading (optional)Measured on the real machineReplaces this article’s 50 kPa assumption and improves the accuracy of the estimateYou

When the physical item arrives at Jiahon, we first do a trial pick and measure the size before issuing the drawing; the nozzle is made after both sides confirm the drawing, and acceptance is by the drawing dimensions. This is the general case, and individual cases still vary.

One item often left out, and very cheap, is the ball map — it decides the recognition gate, and can be judged first without any sample part. Another often misunderstood is speed: many times the part is not impossible to pick up, but the line wants to move a heavy part with the G value of an ordinary part; settle the acceptable cycle time first, and the nozzle size becomes much more reasonable.

For standard items, the Panasonic-fit NOZZLE 226 / 230 / 235 / 240CS and Panasonic-fit new nozzle series, commonly used on the Panasonic NPM series, are standard specifications; for Yamaha, FUJI and JUKI machines see YAMAHA nozzles, FUJI nozzles and JUKI nozzles. Custom items are quoted after evaluation of the actual part.

› Related products: Nozzle customization (odd-shaped nozzles)

Need help judging?

Tell us the part and the head type, and let Jiahon work through the four gates for you

Give us the model and head type, the part size and weight, and a bottom view or a photo of the underside. Using the method in this article, we will reply with where this part falls at each of the four gates of pick, move, recognize and place, explain how far the nozzle end can go, and which items must be confirmed with the machine manufacturer. Send the physical item to Jiahon and a trial pick can be arranged.

Sources and notes

  • Manufacturer public data (all viewed 2026-09-19): Panasonic NPM-W2 (catalog 2024-05-31), NPM-WX (catalog 2022-04-25), NPM-GW (catalog 2025-01-16) and the Offline Camera Unit V2 product page; ASMPT, Technical Overview – AI Server Applications, Edition 01/05-2026; FUJI news release “Fuji Enables SMT Placement of Large, Heavy Components for AI Servers” (2026-07-30) and head product pages; Yamaha YRM20DL product page; JUKI JM-E01 product page and RS-2 catalog (2024-12); Universal Instruments Fuzion OF datasheet MC-7490; Hanwha XM520 product page; Mycronic MYPro A41 datasheet (2025-01).

  • Formulas and values put in: Schmalz, “Theoretical Holding Force of a Suction Cup”, load cases I / II; Schmalz, “Design of the Suction Cup” (safety factor: 1.5 or more for smooth workpieces, 2.0 or more for critical, rough or oily workpieces; friction coefficient 0.5 for metal, glass and wood); SMC vacuum pad selection guide (safety rate: 4 or more for horizontal lifting, 8 or more for vertical lifting, a static calculation with no separate acceleration); SMC ZM series vacuum generator catalog (vacuum upper limit −84 kPa for the standard model at 0.5 MPa supply, and the vacuum value falls as leakage increases). This article takes ΔP 50 kPa, μ 0.5, S 2.5, a 1–3 G; the weights in Table 1 are each manufacturer’s published limit.

  • For the explanation that a regular array cannot be located alone, see the public content of Cognex US patent US6151406A and KLA US patent US11113827B2.

  • Each manufacturer’s values were obtained under that manufacturer’s own mechanism and control conditions and must not be used to infer the feasibility of another brand’s machine; the copyright of manufacturers’ data belongs to each manufacturer, this article cites it as a source, and the data follow each manufacturer’s current public version.

  • The Jiahon affiliated company does not sell or act as agent for any brand of mounter; “fit” in the text means supplied items matching in size and specification, not parts made by the equipment manufacturer.

  • All calculated values in this article are engineering estimates, not promised values of Jiahon or any manufacturer; for actual adoption recalculate with values measured on the real machine. This is the general case, and individual cases still vary.

Back to top