Wire Bonding and OSAT Packaging: The 25 µm Gold Thread, the ₹32 of Gold in a 100-Lead Package, and the 3 gf Pull Test Behind India's Chip Plants

Engineering guide to wire bonding and semiconductor packaging: free-air ball volume conservation, 25 µm gold vs copper wire, Au-Al and Cu-Al intermetallic growth, MIL-STD-883 pull and JESD22-B116 shear limits, mould compound CTE and warpage, J-STD-020 moisture levels, the ₹0.32 of gold per wire, and India's 2026 OSAT plants.

·

Wire Bonding and OSAT Packaging: The 25 µm Gold Thread, the ₹32 of Gold in a 100-Lead Package, and the 3 gf Pull Test Behind India's Chip Plants

On 28 February 2026, Micron's assembly and test plant at Sanand was inaugurated. Kaynes Semicon followed on 31 March, and CG Semi — CG Power with Renesas and Stars Microelectronics — started commercial packaging on 4 July. As of November 2026, all three of Sanand's announced OSAT and ATMP plants are shipping product, Tata's ₹27,120 crore plant at Jagiroad has begun pilot packaging, and the Dholera wafer fab is still a construction site with commercial output now guided to mid-2028.

None of those three plants makes silicon. They take wafers fabricated elsewhere, thin them, dice them, glue a die into a package, connect the die to the leadframe, encapsulate it in epoxy, and test it. The connection step is wire bonding, and it is the least glamorous, highest-volume, most consequential operation in the entire semiconductor supply chain. A modern ball bonder places a 25 µm gold or copper wire at 12–16 wires per second with ±2.5 µm placement accuracy. The bond itself forms in 10–20 ms through 120 kHz of ultrasonics applied at 20–50 gf, and it is accepted or rejected on a pull test whose minimum for 25 µm gold is 3.0 gf — a force roughly equal to the weight of a small paperclip.

This is the engineering inside that number.

What an OSAT actually does

The back-end flow is short in description and brutal in detail. Each step has a process window, and each window interacts with the next.

Step · What happens · Numbers that matter

Backgrind · Wafer thinned from ~775 µm · 75–100 µm for thin packages, 50 µm for stacked die; TTV < 2 µm

Dicing · Blade cuts die apart · 30–50 µm kerf, 40,000–60,000 rpm, 30–50 mm/s feed; chipping < 10 µm

Die attach · Silver-filled epoxy or sinter · 60–80 wt% Ag epoxy, 10–25 µm bond line, < 5% void, cure 150–175 °C for 1–2 h

Plasma clean · O₂/Ar activation before bonding · 2–5 min; contact angle < 20°

Wire bond · Die pad to leadframe/substrate · 25 µm wire, 16 wires/s, 63 ms cycle, 120 °C stage

Mould · Transfer-mould epoxy encapsulation · 175 °C, 60–90 s, 40–85 kg/cm², 86–90 wt% silica filler

Post-mould cure · Cross-link the epoxy · 4–6 h at 175 °C; Tg rises to 150–180 °C

Trim/form · Singulate, bend leads · QFN sawn with 0.2 mm blade; coplanarity ≤ 0.1 mm

Plating · Matte Sn or NiPdAu finish · 5–15 µm Sn; whisker risk drives NiPdAu

Test · ATE at temperature · 8–32 sites, 0.5–10 s per unit, −40 to +125 °C

Burn-in · Accelerated early-life removal · 48 h at 125 °C for automotive

Tape and reel · Ship · 2,000–5,000 units per reel

The wire bond step sits between two cleans and before the mould. Everything downstream — mould flow, thermal cycling, moisture sensitivity — stresses the bonds that were made in those 63 ms. That is why the process window is narrow and why the inspection is destructive.

The bond cycle: 63 milliseconds, six events

A ball bonder repeats the same sequence for every wire:

  1. Electronic flame-off (EFO). A high-voltage spark jumps a 0.5–1 mm gap to the wire tail, melting its tip. Typical settings are 4,000–6,000 V, 20–35 mA and 0.5–1 ms. Surface tension pulls the melt into a sphere before it solidifies.
  2. Free-air ball (FAB) formation. The ball diameter lands at 1.8–2.5× the wire diameter — roughly 45–50 µm for 25 µm gold wire.
  3. Impact. The capillary descends, the ball contacts the aluminium pad at 20–50 gf, and the ball squashes. Contact velocity and impact force determine how much of the ball spreads before ultrasonics starts.
  4. Ultrasonic bonding. The transducer drives the capillary at 60–120 kHz. Friction at the ball–pad interface breaks the native Al₂O₃, asperities weld, and an Au-Al intermetallic layer nucleates. Bond time is 10–20 ms.
  5. Looping. The capillary rises to the loop height, pays out wire, and travels to the second bond site. Loop shape is controlled by the wire feed and the capillary trajectory.
  6. Second bond and tail break. A wedge (stitch) bond is made on the leadframe or substrate pad, then the wire is pulled and broken below the heat-affected zone, leaving a tail for the next EFO.

Volume conservation sets the ball geometry

The FAB has to become a bonded ball of a specific diameter and height. Volume is conserved to within a few per cent, which gives the process engineer a hard constraint:

For a 25 µm wire with a 45 µm FAB:

A bonded ball of 60 µm diameter and 18 µm height contains:

The 6.7% excess is the fillet that squeezes out beyond the capillary chamfer. It is also the reason the ball-to-wire diameter ratio is not a free choice: at a fixed FAB diameter, a flatter ball must be wider, and a 60 µm ball at 18 µm height gives a diameter-to-height ratio of 3.3 — right in the 3–5 band that most fine-pitch processes target. The equal-volume diameter at 18 µm height is 58.1 µm, so the 60 µm target is deliberately slightly over-filled to guarantee a continuous fillet.

The capillary constrains the geometry from the other side. A typical fine-pitch capillary for 20–25 µm wire has a 24–30 µm hole, a 29–35 µm chamfer diameter, a 70° chamfer angle and an 8° face angle. The hole must be 1.2–1.3× the wire diameter: too small and the wire drags, too large and the ball sits too deep in the chamfer and the neck becomes irregular.

What ultrasonics actually does

The bond does not form by melting. It forms by solid-state welding under combined pressure, temperature and 60–120 kHz vibration. The vibration imposes cyclic shear at the interface; the aluminium oxide breaks up, fresh metal contacts fresh metal, and the two lattices interdiffuse. The interfacial temperature rises by 200–400 °C for microseconds at asperity contacts, but the bulk ball stays near the stage temperature.

Ultrasonic energy also softens the metal — the Blaha effect — reducing the yield strength of both the ball and the pad during the bond, which is why a 20 gf force can deform a ball that would otherwise need several times that load. Too little ultrasonic energy gives an incomplete bond (small bonded area, low shear); too much gives pad cratering, aluminium splash and a weakened neck.

The heat-affected zone (HAZ) left by EFO extends 30–100 µm up the wire. It is softer and coarser-grained than the drawn wire, and it is where the neck breaks if the loop is pulled too hard. Low-loop processes want a short HAZ; high EFO current with a short firing time is the standard lever.

Four numbers that define the process window

Wire bond recipes are machine-specific — one bonder's "80" is another's "120" — but the physical variables are universal. The table below is the shape of the window, not a recipe.

Parameter · 25 µm Au (thermosonic ball) · 25 µm Cu (ball, N₂) · 300 µm Al (wedge)

Stage temperature · 120–150 °C · 150–175 °C · 25 °C (room)

Bond force · 20–50 gf · 40–80 gf · 300–1,000 gf

Ultrasonic power · 60–120 mW · 80–150 mW · 1–3 W

Bond time · 10–20 ms · 15–30 ms · 20–50 ms

Atmosphere · Air · N₂ or 95/5 N₂–H₂ · Air

Pad metallisation · Al–0.5Cu, 0.8–1.5 µm · Al–0.5Cu or Cu · Al or Cu, thick

Two entries deserve emphasis. Copper needs a shielding atmosphere at EFO: bare copper oxidises within milliseconds at 1,000 °C, and an oxidised FAB produces non-stick-on-pad failures and cratering. Forming gas — 95% N₂ with 5% H₂ — reduces the oxide and gives a smoother ball; pure N₂ works at coarser pitch but is marginal at fine pitch. The aluminium wedge process is not a heavier version of gold ball bonding: it is a room-temperature process for power devices, where a 300 µm wire carries 15–25 A and Al-Al bonds avoid the Au-Al intermetallic entirely.

Gold, copper, palladium: the materials decision

The intermetallic is the reliability clock

Gold ball bonds on aluminium pads form Au-Al intermetallics: Au₅Al₂, Au₂Al, AuAl and AuAl₂. AuAl₂ is the purple phase that gave "purple plague" its name, but the colour is cosmetic. The failure mechanism is Kirkendall voiding. Gold diffuses into aluminium faster than aluminium diffuses into gold, so vacancies accumulate on the gold side of the interface and coalesce into voids. After enough time at temperature the bond lifts with a void sheet underneath it.

Intermetallic growth is diffusion-controlled:

Q for Au-Al is roughly 0.7–1.0 eV, which is why a bond that survives 1,000 h at 150 °C can fail after a few hundred hours at 175 °C. The acceleration is not linear; the Arrhenius term is the whole story.

Copper on aluminium grows Cu₉Al₄ and CuAl₂ at roughly one-fifth the rate of Au-Al, with a higher activation energy. That is the reliability argument for copper: less Kirkendall voiding, better high-temperature storage performance, and a bond that does not care as much about a 175 °C ambient. The counter-argument is corrosion. Cu-Al intermetallics are susceptible to halide attack; chloride or bromide from the mould compound, flux residues or contaminated plating hydrolyses the intermetallic and cracks it. Copper wire qualification therefore controls mould-compound halide content, pH and delamination, and automotive customers require AEC-Q006 rather than the older gold-oriented qualification.

Silver wire sits between the two: Ag-Al intermetallics grow slowly and silver has the highest conductivity, but it tarnishes with sulphur and migrates under bias and humidity.

Palladium-coated copper (PdCu) is the practical compromise for fine pitch. A 0.5–1.5 µm palladium shell protects the wire from oxidation, extends shelf life, and gives a more spherical FAB. It costs 10–20% more than bare copper and is still far below gold.

The cost arithmetic, with November 2026 prices

Take 25 µm wire, a 2.5 mm loop, and 24K gold at ₹13,700/g.

A 2.5 mm wire therefore contains 23.7 µg of gold:

A 100-lead package with one 2.5 mm wire per lead carries ₹32.45 of gold. Every million such packages consumes ₹3.2 crore of gold before any processing. This is the single most important number in package cost engineering, and it is why gold wire disappeared from consumer devices.

Copper at \13,610/t and ₹94.3/\ is ₹1,283/kg, or ₹1.28/g. The same 2.5 mm of 25 µm copper contains 11.0 µg and costs ₹0.000014 — a metal-cost ratio of roughly 23,000:1.

The plant-level saving is smaller than that ratio suggests. Published conversion studies report 63% material-cost savings at 20 µm and about 90% at 50 µm, because the comparison is between drawn, cleaned, spooled bonding wire and drawn, palladium-coated copper — not between two lumps of metal. Copper also brings forming gas, a tighter process window, lower bonder throughput on hard pads, and a yield risk that can eat the wire saving on a high-I/O device. The rule most packaging engineers use is blunt: copper wins above 50 µm and on mature nodes with robust pads; gold or PdCu wins at fine pitch, on low-k dielectric stacks, and wherever the qualification cost of changing the wire exceeds the material saving.

The wire as a circuit element

A bond wire is not an ideal short. Its resistance follows the usual relation:

For 2.5 mm of 25 µm gold (\rho = 2.3 \times 10^{-8}\ \Omega m), R = 117\ \text{m}\Omega. The same length of copper (\rho = 1.68 \times 10^{-8}\ \Omega m) is 86 mΩ. At 1 A that difference is 31 mW per wire — irrelevant for a signal, decisive for a power device with sixteen parallel wires.

Current capacity is set by fusing. The Preece equation is the classical estimate:

with D in inches and k = 10{,}244 for gold or copper in free air. For 25 µm gold, D = 9.84 \times 10^{-4} in, so I \approx 0.32 A in free air. Encapsulated in mould compound, the modified Preece constant rises to 30,000 for wires shorter than 0.040 in, giving about 0.93 A. A supplier datasheet for 10 mm of 25 µm gold quotes 0.52 A, between the two — which is exactly the point: the number depends on length, encapsulation and allowable temperature rise, and the datasheet value is not a universal rating.

Inductance matters as much as resistance at frequency. A 25 µm wire loop runs at roughly 0.7–1 nH per mm, so a 2.5 mm loop is about 2 nH. At 1 GHz that is 12.6 Ω of reactance — which is why RF and high-speed digital designs move to ribbon bonds, flip chip or copper pillars rather than trying to shorten wire loops.

Power modules invert the geometry. An intelligent power module or an IGBT half-bridge uses 125–500 µm aluminium wire or aluminium ribbon, wedge-bonded at room temperature, carrying 15–100 A per connection. Kaynes Semicon's Sanand line, which started with intelligent power modules, is built around exactly this heavy-wire process rather than 25 µm gold ball bonding. The physics is the same ultrasonics; the scale is a different machine.

Loop control and wire sweep

The loop is a structural element. It must clear the die edge, stay below the mould cap, and survive the moulding compound flowing past it at 0.1–1 m/s. Standard loop heights are 200–250 µm; low-loop processes run 75–150 µm for thin packages and stacked die.

Wire sweep is the moulding defect that kills fine-pitch parts. The melt front drags on each wire; a long, high loop in a fast-flowing region sweeps downstream and can short against its neighbour. Most automotive customers cap sweep at 5–10% of the loop span. Copper helps here — its modulus is 110–130 GPa against gold's 78 GPa, so the same drag produces less deflection — and so does reverse bonding, where a stud bump is placed first and the wire is bonded stitch-first to keep the loop low and steep.

The design rules that follow are absolute: no wire crossings, wire-to-wire spacing of at least one diameter, loops aligned with mould flow, and a gate that does not put the highest loops in the fastest melt.

The package is a thermo-mechanical joint

Encapsulation is where the electrical design meets thermal expansion. The relevant coefficients:

Material · CTE (ppm/°C) · Young's modulus

Silicon · 2.6 · 130–170 GPa

Copper leadframe (C194) · 17.6 · 120 GPa

Alloy 42 (FeNi42) · 4.5–5.3 · 140 GPa

BT substrate · 15–17 (x-y) · 20–25 GPa

Mould compound below Tg · 7–8 · 22 GPa

Mould compound above Tg · 27–45 · 0.45 GPa at 260 °C

The mismatch strain between die and mould compound over a 150 K excursion is:

For a 5 mm die that is 4.05 µm of relative displacement, and if the compound is mechanically constrained the stress is:

Twenty-five megapascals across a mould compound–die interface is enough to delaminate a poorly adhered surface, and delamination is the precursor to every moisture-related failure. The compound is therefore 86–90 wt% spherical silica — silica has a CTE of about 0.5 ppm/°C, so filler loading is the main lever on CTE — with a low-stress biphenyl or OCN resin, a moulded shrinkage of 0.2–0.3%, and a spiral flow of 127–132 cm at 175 °C to fill a multi-cavity mould before gelation.

Warpage is the same physics at package scale. A QFN moulded at 175 °C cools to 25 °C with the compound on one side and the leadframe on the other; the bimaterial strip bows. JEDEC JESD22-B112 measures warpage at 25 °C and at reflow temperature, and the acceptance limit for a 7×7 QFN is typically tens of micrometres, with lead coplanarity held to 0.1 mm.

Moisture is the last variable. Mould compound absorbs water; at 260 °C reflow that water flashes to steam and the package can popcorn. J-STD-020 defines moisture sensitivity levels 1–6 by floor life, and every package is qualified at its MSL. A package that passes MSL 3 at 260 °C has been baked, soaked, reflowed three times and inspected for delamination by scanning acoustic microscopy.

What is actually tested

Wire bond qualification is destructive by design. The two standard tests are:

The failure mode is more informative than the force. A wire break in the span means the bond is stronger than the wire and is acceptable. A ball lift, pad crater or stitch lift is a process failure regardless of the gram reading. Neck breaks point to EFO or HAZ problems; heel cracks point to loop geometry. That is why the pull test records both the force and the failure code.

Reliability qualification then stresses the parts and repeats the destructive tests:

Test · Condition · Duration

Temperature cycling · −55 to +125 °C · 1,000 cycles

High-temperature storage · 150–175 °C · 1,000 h

HAST · 130 °C / 85% RH · 96–264 h

Moisture sensitivity · J-STD-020, 260 °C reflow · 3 passes

HTOL · 125 °C, biased · 1,000 h

Drop test · JESD22-B111 · Handheld classes

Acceptance is usually no more than 25% degradation from the pre-stress pull and shear distributions, plus zero delamination beyond the J-STD-020 criteria. In production, ball shear and pull are sampled per lot and tracked as Cpk; 1.67 is the usual minimum for a safety-critical automotive part, and 100% automated optical inspection checks ball placement, ball diameter and loop height on every unit.

The economics: ₹0.32 per wire

Put the cost stack together for a 100-lead package at 2.5 mm average wire length:

Cost element · Per unit · Basis

Gold wire · ₹32.45 · 100 × 2.5 mm × 9.47 µg/mm × ₹13,700/g

Bonder time · ₹5.21 · ₹3,000/h fully loaded at 16 wires/s best case

Capillaries and consumables · ₹5.66 · \$0.04–0.08 per 100-wire unit

Variable total · ₹43.31 · Gold is 75%

Copper wire collapses the first line to ₹0.0014 and the total to ₹10.87 — a 75% reduction at the metal level, before yield and throughput penalties. That single table explains the last fifteen years of copper conversion better than any reliability paper.

The capacity arithmetic is equally unforgiving. A 100-lead device needs 6.25 s of bonding at 16 wires/s, plus about 1.5 s of handling — 7.75 s per unit. At 85% uptime, one bonder produces about 9,500 units per day. A plant advertising 6.33 million units per day would need roughly 670 bonders if every unit had 100 leads, and about 1,590 for a 15 million/day line. Those numbers are why OSATs chase low-lead-count, high-volume discretes and power packages, and why the "chips per day" headline is meaningless without the package mix behind it.

Testing is a smaller but not trivial cost. Automated test equipment runs at \150–300 per hour; a 1 s test at 16 sites works out to a few paise per unit for simple analog parts and tens of paise for complex digital ones. Packaging gross margins run 15–25%, testing 25–35%, and packaging is about 75% of the \40–50 billion annual OSAT market — the volume is in packaging, the margin is in test.

What India's OSAT build-out actually changes

Plant · Investment · Capacity (nameplate) · Status, November 2026

Micron ATMP, Sanand · ₹22,516 crore · ~14 million units/week · Commercial since 28 Feb 2026

Kaynes Semicon, Sanand · ₹3,307 crore · 6.33 million chips/day · Commercial since 31 Mar 2026

CG Semi, Sanand · ₹7,584 crore · 15.07 million units/day at full build · Commercial since 4 Jul 2026

Tata Electronics, Jagiroad · ₹27,120 crore · up to 48 million chips/day · Pilot packaging reported Oct 2026

Tata–PSMC fab, Dholera · ₹91,526 crore · 50,000 wafer starts/month · Under construction; mid-2028 target

The strategic point is not the unit counts. It is that packaging is a materials-and-tooling business, and almost none of the materials or tooling is made in India yet. Bonding wire comes from a handful of Japanese, German and Korean drawers. Mould compound comes from Japanese and German formulators. Leadframes are stamped by specialised suppliers. Substrates come from a short list of Asian laminate houses. Capillaries are ceramic parts with a 25 µm bore. Bonders come from Kulicke & Soffa, ASM Pacific and Shinkawa.

That is the opening for fabrication shops. A leadframe stamping die is a progressive carbide tool running at 800–1,200 strokes per minute on 0.1–0.2 mm copper alloy; the die clearance is measured in microns and the tool life in millions of strokes. Mould cavities and blocks are hardened H13 or S136, mirror-polished, and often finished by EDM and hard milling. Test sockets and handler fixtures are machined from PEEK and PEI with pogo pins on 0.3–0.5 mm pitch. Trim-and-form tools, device trays, cleanroom panels and precision fixtures are all conventional precision-manufacturing work with an unusual qualification burden.

That burden is the real barrier: an automotive packaging customer will not accept a new leadframe or mould insert without PPAP, IATF 16949 and 18–24 months of qualification. The opportunity is real, but it is a tooling-and-qualification opportunity, not a "quote and ship" one.

Where wire bonding stops

Wire bonding has a hard physical ceiling at fine pitch. Production has reached 40 µm pad pitch; the in-line capability is about 35 µm. Below that, the bonded ball is wider than the space between pads, wire sweep during moulding exceeds the clearance, and the ultrasonic energy needed to bond craters the low-k dielectric under the pad.

The alternatives each trade one problem for another:

Wire bonding survives because it is the lowest-cost interconnect per I/O and because the installed base is enormous. At 40 µm pitch, with copper wire and a low-loop process, it remains the right answer for most microcontrollers, power devices, sensors and memory packages — precisely the product mix India's new plants are targeting.

The specification checklist

If you are sourcing a packaged device, a leadframe tool or a mould insert, these are the numbers that belong on the RFQ:

None of that is exotic. All of it is what separates a package that survives 1,000 thermal cycles from one that lifts a bond at 600.

The 25 µm wire is the last analogue step in a digital supply chain. It is drawn to a tolerance of a micron, bonded in 20 ms, inspected by pulling it until it breaks, and paid for by the microgram. India's OSAT plants are, at bottom, a bet that the country can run that step at millions of units a day — and a signal that the next layer of domestic manufacturing opportunity sits in the tooling, materials and fixtures that surround it, not only in the cleanroom.

More FabFlow blog posts