Home / Insights / Soft Landing in Die Attach: Why Contact Force Beats Position Accuracy

Technical note

Soft Landing in Die Attach: Why Contact Force Beats Position Accuracy

Die attach is a force problem, not a positioning problem. The parts that get damaged are damaged by contact force, and the specification that protects them is a repeatable contact force rather than a repeatability figure for position. This note reads the contact-force specification on a 48 V DC servo drive honestly, compares it with the 10 g figures published elsewhere in the industry, and sets out how I would qualify a soft landing on your own parts.

The damage is a force problem

The most common way a die-attach machine damages a part is not a positioning error. It is a contact force error. A die that lands a few micrometres off target may still bond; a die that lands exactly on target with too much force may crack, may damage the bond wires on the die behind it, or may mark the pad so the wire bond does not hold. That is why how accurately does it place is the wrong question for this process, and why contact force is the specification that actually governs yield. Position accuracy still matters — you cannot put the die in the right place if you cannot repeat the approach — but it is an enabling requirement, not the protective one.

How to read a 10 g specification

When a datasheet says contact force is typically around 10 grams, with roughly ±1 gram variation, read it precisely. Ten grams is a typical value, not a setpoint the drive holds to on every part and not a number the process inherits for nothing. The ±1 gram describes the variation around that typical figure, and whether your process can live with it depends on the part, the pad stack and the tool. The high-performance -P version offers finer force resolution, 0.1 g, which is a different question from a different number: it asks how finely the current loop can be trimmed, not what the process tolerates. Current sampling is 12-bit as standard, with 16-bit on the -P and -AP variants. What I would not do is read 10 g as a value every part receives, because nobody who has built these machines can promise that — which is why I would rather qualify it on your parts than quote it as a headline.

Force is current, and the current loop is fast

Force is torque is current, and that is the whole argument for a 48 V DC architecture. On these drives the current control frequency is 20 kHz, maximum 40 kHz, and the force loop rides on top of it. The architecture note explains this in more detail, but the engineering consequence is simple: the speed at which a contact event can be caught and corrected is set by the current loop, not by the size of the motor. On a large AC servo platform, or on a voice-coil actuator, the same job is done with different mechanics and a different loop structure, and the published contact-force figures come out of those structures. That is not a criticism of them; it is a reminder that a force number without a mechanism behind it is not a comparable number.

The published numbers in this market

Several established suppliers publish 10 g contact-force figures. MRSI Systems quotes forces as low as 10 grams. SMAC's Soft-Land™ is described as gentle touch at 10 g. At least one competitor publishes finer numbers again. Their 10 g figures come from AC servo platforms and voice-coil actuators. Ours comes from controlling motor current directly on a 48 V DC bus. So if your comparison sheet has three columns reading 10 g, that column separates nothing, and the useful question is the same one I would ask of any supplier: is the force produced by controlling motor current, or by a sensor and a control loop arranged around the tool? I have written more about this in the note on why the 10 g figure is not a differentiator, and the wider process map is in the semiconductor post-bonding solution map.

What the drive contributes to the process

The table below is the part of the specification that touches the process. I have kept it separate from the marketing numbers on purpose, because these are the items your own engineer can check against the datasheet.

The stage numbers that set your cycle time

The stage is where the cycle time lives, and the dual-axis ZR unit is where that work is done. GM1-E48xxN-2D-ZR measures 127.5 × 65 × 28.6 mm and covers the two axes of a ZR stage. Its published figures are rotation repeatability of ±0.01°, linear repeatability of ±2 µm, cycle time below 50 ms per revolution and below 80 ms per 30 mm of linear travel. Those are the numbers that decide whether your cycle budget works, and they are process-relevant rather than marketing-relevant: a 50 ms rotational step is a real constraint on a multi-step bond cycle. At the finer end of the same market, the UF platform's -AP variant is specified at ±30 nm static jitter and ±70 nm repeatability while moving at 300 mm/s. Nanometre placement and gentle landing are separate specifications, and a machine may well need both — they are answered by different parts of the same drive.

Qualifying on your own parts

If you want to know whether these drives land your parts properly, qualify on your own parts rather than on a datasheet, and the sequence is straightforward. Instrument the tool — a load cell on the tool, even temporarily — and log the contact force across a batch of your actual parts at the speed and the deceleration you intend to run. Look at the spread, not the average: an average of 10 g means nothing if the spread is ±3 g. Then repeat the measurement at the dwell you plan to use, because dwell changes the reading when the part deforms or when tool compliance matters. I would also check the behaviour when the part is missing, when it is presented at an angle, and when the previous part left debris — the contact event is where all of those show up. End-of-travel vibration suppression is available as an option rather than standard, so if your machine hits a hard stop at the end of travel, confirm it per order rather than assuming it.

What I need from you

What I would need from you to make this concrete: the part and the pad stack, the tool geometry and its compliance, the force window your process can accept, the cycle time you are trying to hit, the ambient conditions, and whether the machine runs at room temperature or has to survive a wide range — the DG wide-temperature family covers −40 to +70 °C while the UF series is rated 0 to +45 °C, and die-attach lines usually have both a hot section and a cold one. The force control application note is in the document library, the model-level detail is on the model parameters page, and if you would rather discuss the specifics of your process, the contact page is the fastest route.

Key takeaways

• Die-attach yield is governed by contact force repeatability, not by position accuracy; the position spec is an enabling requirement.
• Contact force is typically around 10 grams with roughly ±1 gram variation on a 48 V DC drive, and the -P version offers finer force resolution of 0.1 g.
• Judge a supplier by the mechanism: the 10 g figures published by other suppliers come from AC servo platforms and voice-coil actuators, while this architecture controls motor current directly on the 48 V DC bus.

Common questions

What contact force can the drive hold on a die?

Contact force is typically around 10 grams, with roughly ±1 gram variation. The high-performance -P version offers finer force resolution of 0.1 g. Those are typical values with a stated variation, not an absolute figure every part receives, which is why I recommend qualifying on your own parts.

Why do several other suppliers also quote 10 g?

Because the figure is a common result in this market rather than a rare one. MRSI quotes forces as low as 10 grams and SMAC describes Soft-Land™ as gentle touch at 10 g; those figures come from AC servo platforms and voice-coil actuators, while here the force comes from controlling motor current directly on a 48 V DC bus. The mechanism is the meaningful difference.

Do I need a separate force sensor to qualify the landing?

For production validation I would instrument the tool temporarily — a load cell on the tool gives you the batch data you need to judge spread rather than average. The drive's own probe function handles contact detection and force feedback in the machine, and the force control application note explains the behaviour on your workpiece.