Copper Wire Bonding

I have been working on copper wire bonding for a while and I must say that this is the best thing invented so far that beats using gold wire bonding. The differences are vast due to the fact that copper wires oxidizes easily and even when coated with Palladium, they have a tendency not to create a rounded free air ball after a certain period of floor life staging.

I have only seen memory makers shunning from using copper wire bonding so far. The study that was conducted has so far shown that problems would not occur up to DDR2 and I suspect should not even be a problem for DDR3 and DDR4. Looking at the characteristics of copper wire, the conductivity is much better with lower resistivity. Reliability wise is also much better due to the slower growth of IMC.

Of course, there is also the case of corrosion on copper wire bonding due to flourine and chlorine. The molding compound used so far helps to reduce this corrosion problem. What is your say about copper wire bonding?

GR&R for development?

I heard a funny request from my supervisor. He had wanted to perform a GR&R in development stage for any devices that are under qualification. I find this a bit wrong from a GR&R standpoint.

I believe that GR&R are equipment based and should be done for any equipment in a 6 or 12 month period to ensure that

1. Equipment are still able to measure accurately
2. The people handling these measurements are still well trained and have a standard set of measurement techniques

Why I say this is funny is wrong to perform for every device is that, we are effectively still using the same tools for measurement. It does not make any sense to perform it every time a product comes into development if the same set of tools are used. I would agree if it is development using a new set of purchased tools/equipments.

Do let me know your views and comments on this.

Wirebond NSOP - "M9" device

Looking at a poorly characterized process for this particular device with 4um Al bond pad thickness and frequent NSOP >200PPM, this device could barely run without having the operator assist every few minutes. If left unresolved in HVM, this would exponentially increase the operator's workload which in turn would cause operators to mis-process steps as had been proven in the past. The key is to always ensure operators are always at a low level of stress so that they are fully concentrated on the correct way of doing things and not do things in a pressured manner.

First of all, pulling yourself out of the picture, from the start, this device was only qualified to bond with good reliability results and nothing on the running performance and definitely not a large enough sample size to verify yield and bondability performance. In characterization, one would always need to includ alarm signals then only can it be claimed as a full characterization.

Now, looking at the pressure the higher management are in to ensure this device performs, I'd have to use some theory to derive a solution based on the observations in the quickest manner coupled with a short DOE. Please do not try this if you are unfamiliar of what are the relationships of the parameters with the motion of the head. Always understand what it is you want to achieve and what are the associated parameters that can help you before you jump into adjusting every parameter you see. This cannot be taken lightly I have seen others venture into parameters not knowing their functions causing further problems.

First, collect information on the NSOP and see if the NSOP are localized or random, in this case random. Is the ball size within the expected bond pad and is the BAR ~3.5-4.5 is my first rule of thumb. Collect also information on the ball shear and shear/area while always trying to target >8.5 shear / area - in this case it is. All these information tells me that the parameters are likely at the process borderline and what needs to be done is to shift the process upwards to a higher shear value. Collecting the NSOP alarm data and the shear value will tell you this exactly. Blindly increasing USG does not help due to device being Cu wire based and causes excessive aluminium splash on increase of USG. So to go about this, I have put up a DOE for initial force to be a smaller range (8-10) of the base parameter's force (12) and also reducing the bond time from 15 to 12 while increasing the USG range from 78 to 85.

The explanation behind what was done above is to first create an initial force to squeeze away some of the thick aluminium to prepare the surface for bonding before actual base parameters. Since the interdifussion is largely dependent on the USG, it has to be increased to improve the shear effectively. Hence to reduce the aluminium splash from the increase of the USG, the bond time had to be decreased. Of course, a full blown DOE would beable to tell you this but in this situation I was short of time. Indeed the short DOE had shown that this method compensated for the increase in USG by reducing the time while maintaining splash and ball size. In order to verify the parameters derived from the maximum desirability, of course the  a measureable aluminium splash, ball shear, shear/area and ball size results with t-test comparison with POR is required. This device being a 4um Al thickness, I worry less about the cratering but of course should always be part of your assessment.