Continued variability in the performance of diamond
conditioners shows that the suppliers lack an effective quality control (QC)
tool. What is required for an effective diamond conditioner QC tool?
The most systematic work I've seen on this topic has come from Araca. Until there is a body of first-order knowledge accessible to and used by all players, empirical functional testing will remain the standard method for QC sampling. However, this approach will become more statistically significant as the conditioners themselves migrate toward micro-replication manufacturing methods rather than random distribution and orientation of cutting edges.
I agree with Mike. There is limited understanding of key parameters impacting conditioner performance, which in turn limits the tools that can be deployed for effective QC. Another aspect is diamonds themselves.
As an example, geometry of cutting tool/tip used in machining has significant impact on quality of cut, and very complex shapes have been developed to optimize cutting efficacy; while diamond shapes for conditioners are very coarsely defined.
I would agree that microreplicated diamond tips or monodispersed diamonds with identical shape in a structured array would be a good start.
Current QC methods used by diamond conditioners do not necessarily mimic actual use conditions during CMP processes. They are required to montior/maintain current state of variability but they are not sufficient. For example there are various techniques used by suppliers for diamond pull-out, adhesion between diamond grains and substrate but these methods do not mimic actual CMP conditions. Some are done on individual diamonds (so they lack statistical correlations). Any new method development should mimic actual CMP conditions (e.g., done on fully manufactured disk, expsoure to chemical enviroment during CMP, with disk rotation, etc.). Some recent work by UofA/Araca is promising and step in the right direction.
The fact is that there is no standard method to evaluate conditioning abrasives. Until there is a hint of a standard, the metrology approaches will be all over the map and not consistent. Like other processes in semiconductor manufacturing (where standard test methods do not exist), there may be too many parameters to allow development of a standard testing method (conditioner, tool, slurry, pad, and process conditions). As Mansour said for the post-CMP cleaner approach - the end user will use what works. QED
Within a batch and batch-to-batch diamond disc quality (as measured by pad cut rate as well as the number of active diamonds) needs much improvement. Based on my observations, pad cut rate within a batch of discs, can often vary by more than 50 percent. Some data I have seen shows cut rates ranging from as low as 10 microns per hour to 70 microns per hour for the same type of disc. This is a major issue since pad cut rate directly impacts CMP performance and as disc life can be quite variable. Regarding the number of active diamonds (these are the diamonds that do the actual cutting of the pad), our tests have shown that only about 0.1 percent of all diamonds on a disc can be considered to be active (somewhere between 100 to 500 diamonds depending on the type of the disc). This fact in and of itself is quite significant since most of the work of pad cutting is done by only a few diamonds. More importantly, the number of active diamonds can decrease by as much as 50 percent over a 20 hour period (this, along with diamond tip micro-wear, is what leads to end-of-life for discs). Therefore, it is critical for disc makers to report the number (and if possible, the location) of the active diamonds on their discs as a QC parameter. Another key consideration is the extent of metal leaching from the discs at working temperatures (i.e. 50 degrees Celsius).
I thought I'd share with you the attached JECS joint Intel-UA-Araca paper that was published electronically last week. I'd love to get your comments about our findings as well as your thougts on what else we should look at.
There is plenty of data indicating that active diamond microwear, say after 30 - 50 hours of operation, is no more than 5 microns. That is, the active diamond edges get rounded off and their protrusion decreases by only 5 microns during this time.
As such, the planarity of the diamond disc substrate (metal of platic or whatever) must be kept to within 5 micons total over the 100 cm diameter of the disc.
I am curious to know whether diamond disc makers are paying attention to the substrate planarity and whether this is a QC parameter.
Ara, unfortunately, not too many people had a chance to attend my talk on effects associated with conditioning action. Briefly, some of the conclusions there:
You are absolutely right that only a very small portion of the diamonds are engaged in the action. It depends on two main factors: a) the global disc profile, and b) the diamond height distribution. The a) has to match the pad contacting profile, which depends on the conditioning down force and type of the pad stack. For example if the disc is flat only the edge diamonds are involved in the action and wear fast, after which the disc is dead because these dull diamonds resting against the pad do not allow others to engage. We found a way to scan the cond-disc/pad engagement profile and ordered customized discs from one of the vendors, and immediate have found practically even engagement across the cond_disc.
The other huge problem is diamond height distribution. The immersion depth of a diamond disk at some reasonable down forces is in the range of 10 um. So, if the diamond height distribution width (95%) is twice larger (20nm), than about 50% of the diamonds are doing their job. However if it is already 40-50 um, then only just a very few are actually acting. The ugly stuff here is that these diamonds represent the tail of the distribution which is just impossible to control. BTW, the presentation was named " The good, the bad and the ugly of the pad conditioning action" and it was presented recently at the CMP UG.
I put together a matrix of the requirements which have to be met to fabricate a good, maybe better to say a right cond_disc, and with these customized discs we got much better performance parameters, process stability, and, what made me especially glad, that diamond wear was uniformly distributed across the disc, and a much larger percentage of them were indeed involved in the action.
As I've already mentioned in one of my messages, that was done quite a while ago (~6 years), and, unfortunately, I see that this is still an issue.
We also developed a procedure, simple one) how to control the diamond wear, and its distribution across the disc, and some other things. I believe, I placed this stuff at your site sometimes last year, but maybe I am mistaken.
Perhaps I'm displaying my ignorance in this followup question, but isn't the major issue with conditioners the desire to keep this consumable cost near historic lows? I ask because ~10 years ago I was involved at AMAT's repairs-and-spares group in working with 3M on a great new conditioner that seemingly exceeded all expectations except for cost...and so the industry merely pushed what it had as much as possible to keep net consumables costs down.
Just for the sake of argument, what if a novel diamond conditioner cost 3X per unit but demonstrated 10X performance (based on improved morphology and height distribution, substrate retention, etc.)...would anyone go for it? Or would everyone merely use this new entrant product as an excuse to push current suppliers to improve specs. (so as to eventually pay 1.2X more to get 1.5X greater function)?
The average selling price for most state-of-the-art discs in HVM is as low at 100 USD (sometimes even near 90 USD). This is a five fold decrease from 7 years ago. Unfortunately, since IC makers have been squeezing the suppliers for lower and lower pricing, the suppliers have not been able to afford implementing new QC procedures and standardization efforts. That's why the quality continues to be horrible.
If a new disc with 3X or 5X of the current HVM price were to make an entrance, it needs to demonstrate (covincingly) a major yield die improvement. Since disc makers don't have the bandwidth to demonstrate yield gains, the only way for them is to co-develop their product with major IC makers. Only then, can they get the higher price for their new product. I think it is entirely possible to ask for 400 USD per unit if yields are better.
Can we look at this in a different light? Is the extent of conditioning current used necessary if a pad management strategy was implemented? Using the conditioner as it currently is used is not only for conditioning the pad but to move junk around and mix fresh slurry with debris. If it was not necessary to move junk (using the conditioning disk) and mix fresh slurry, the conditioner life could be greatly extended, as well as pad life, and both would lower the CoC. A higher unit cost for a disk would be diluted by its longer lifetime and may allow improvements in conditioner design. Improved design and performance is not free, as much as end users would like it to be.
I totally agree with you. Unfortunately, in CMP we continue to clean up the mess caused by the pad through means other than holding the pad maker responsible for their product. I guess that's what happens when there is not enough competition in the pad space, although that seems to be changing.
By my last count, there were 13 reputable diamond disc makers in the CMP space and maybe that's why average selling prices are dropping like a rock.
If the diamonds were used to simply rejuvinate the pad surface asperities (their primary reason for existence), then they would certainly live longer (maybe 2X longer). Unfortunately, the as-received pads need to be broken-in and some of them also come with a skin layer that needs to be removed.
The pad makers have always relied on the enduser to break-in and season the pad surface using diamond discs. This puts unncessary burden on the discs and reduce polisher utilization. Whay are we then paying the pad makers 1000 plus dollars for their product?
What the pad makers should provide (free of charge) is a pad that is already broken in prior to shipment via surface skinning or sand blastoing or some other means.
Ara, I strongly agree with you that the process of pad breaking-in and it QUAL (PB-iQ)is one of the most painful and time consuming steps, definitely requiring solution. However I do not se how it could be done at the vendor's site because this process is machine/process/slurry/cond_device sensitive. And, what is even more problematic, is how they will adhere/un-adhere the pads after breaking-in. And also, I am not sure that it will finally satisfy the most of the end-users. They will be doing their own PB-iQ and qual anyway: nobody will be willing to put under risk production wafers by running the pads without qual.
What I would do in this case, as equipment vendor, is make the platens consisting of two parts (I have a patent on this with LAM): the lower stationary and the upper thinner plate removable. Also a specialized simplified machine just to do the PBiQs, will be placed in the planarization bay. So, once the pad is PBiQ-ed at this machine, the top plate is transfered together with the adhered to it QUAL-ed new pad to the production machine, which will make the machine down time practically negligible. To me it is a much more efficient way to run the planarization bay, the pads will be PB-iQ-ed under conditions best fitting the process ones, it will not be done within the wafer processing space etc. End-users may like it very much. Remembering my production experience well, I would...
Yehiel ... Thx for your comments re: No. 13. Good idea indeed. If the pad makers cannot help in pre-break-in, then the selling price should be reduced proportionately to allow IC makers to implement your patented idea at a net zero cost.
I know some of you young folks won't remember 8 track tapes, but you might remember your grandfathers talking about video tapes, the ubiquitous VHS devices we used before TiVo and the internet. One of the hot selling items from that era was a tape rewinder that did nothing but rewind VHS tapes so that you didn't have to wear out your VCR doing such a mundane, time consuming task. (Minutes count when you're trying to move quickly from Dallas to Miami Vice...)
It just might be feasible to develop an on-site pad conditioner machine that does nothing but pad break in. The preferred mode might use a vacuum chuck so that the adhesive could be left intact for the polisher. A quick review of the empirical data at hand says that for all of the griping about pad break in, no one has ever seen fit to develop such a machine, and to overcome all of the reasons that 'it can't be done.' So, it is an issue because it keeps coming up, but I question whether it's enough of an issue for anyone to actually do anything about it.
The defects related to pad composition and conditioning are much more critical problems, since they affect yield and throughput (OK, break in affects throughput too...) and CoO, but many people are actually doing something about those besides complaining.
In general, to improve our understanding of different pad conditioners relative performance and their variability within a lot as well as lot to lot, it would be essential to perform extensive conditioner characterization work (if feasible, with industry-wide collaboration) employing a number of pad conditioners (varying degree of diamond sizes, shapes, and distribution density; measured pre- and post-test diamond heights, active surface/diamonds %, etc.) and pads for a range of contact pressures and slurry flow conditions. Again, to limit the cost of such characterization, it may be useful to perform benchtop studies first and identify the areas to concentrate for the real-application studies in fab settings, with potential of generating more complete data set with advanced metrology. In the recent past, some efforts were made in this direction, especially in the development of a “draft” test protocol for pad conditioner comparative evaluation. I have attached a working/draft document for reference (courtesy of CETR, Entegris, and other). This “draft” protocol is just an example and a starting point for the development of similar guidelines for characterization of other CMP consumables. Just my 2 cents.
Ara, I doubt that pad makers would provide pre-broken in pads since the break in period is thought to be very consumable and process dependent. If it is not, and the only function of the pad break in is to remove the skin of the pad (which has a density gradient) to provide a greater depth of uniform density or other bulk property, then you have a very good point. Sand blasting or some other abrasion technique outside of the CMP tool could remove the pad skin which apparently is not useful for CMP.
Of the 120 replies posted in the 2010 CMP Roundtable, more than a third dealth with issues surrounding diamond discs and conditioning during CMP.
This is interesting and quite telling. It is also about time that we focus our attention on this often-neglected consumable. Disc makers need to take this very seriously. Top IC makers will be willing to pay more for quality and performance. I know that some disc makers are beginning to take quality seriously, but others are totally clueless.
Also some supplier consolidation will be good (there are more than 13 reputable disc suppliers to the CMP industry).
While for the past 20 years, it had been the other way around, pretty soon, I think that particular diamond disc selections by endusers will affect and even dictate subsequent pad and slurry selections.
We're now (January 20th, 2010) past the
official ending (the 18th) of this virtual roundtable discussion, after 4695 views of 129 replies to 18 questions. I'll
edit together Interesting discussions from most of the topics into a
summary document that will be posted to the Planarization Lounge.
We'll leave the topic posting open in case there are additional comments...but they would not be included in the summary.
Print
One of the hot selling items from that era was a tape rewinder that did nothing but rewind VHS tapes so that you didn't have to wear out your VCR doing such a mundane, time consuming task. (Minutes count when you're trying to move quickly from Dallas to Miami Vice
...)
) and CoO, but many people are actually doing something about those besides complaining.
