Recent studies show that pad debris generated during
conditioning may result in significant CMP defects. How dominant is this effect,
and what new approaches might we take to conditioning or pad materials?
There is both anecdotal and published evidence of pad debris as a defect source at 45nm and below. Its 'dominance' may be a function of how well other defect sources are controlled. The most direct approach for managing it is that taken by Confluense LLC (www.confluense.com) which uses the pad conditioner mechanism to vacuum up pad debris and used slurry as it is created. Some data is available in the publications posted on their website.
Mike, I beg to differ with you regarding the utility of the Confluense device. While they claim that the device reduces defects because pad debris is vacuumed up, it is entirely possible to reduce pad debris during polishing by simply using a better conditioner.
Prof. Park (HanyangUniversity) showed this to be possible in a presentation he gave at the 2009 Lake Placid CMP meeting. Our team (at UA) has also shown that using different conditioners affect pad debris generation by as much as 2X (we are in the process of making the findings public).
There are actually several weaknesses of the Confluense device as follows:
(1) The area that it vacuums is the size of the conditioner, so on every sweep it vacuums up the conditioning debris that it produces but then misses much of the used slurry on the pad.
(2) If the device is used to apply slurry, then it similarly applies it only to an area the size of the conditioner, not the size of the wafer.
(3) The device is designed to work with a TBW conditioner constructed on a substrate that has a hexagonal close packed array of holes (the holes are really necessary). Other conditioners will not work unless holes are provided in their working faces. I don’t think anyone (end-user, tool maker or disc maker) will be willing to do this.
(4) Having an end-user being locked into using a TBW disc is not a very good idea since it is not a high-quality mainstream product compared to the 3M, Morgan Advanced Ceramics, ATI, Ehwa, Saesol and the like.
(5) The Confluense device is very complex and expensive.
Couple of years ago, I sat through a presentation by JSR at the Lake Placid CMP Conference. They showed pretty convincingly that debris generated during conditioning did have a significant impact on wafer level defects. They proved this by 1st collecting and segregating the debris from polishing and then adding it to fresh slurry followed by polishing by this 'debris spiked slurry'.
The only thing that somewhat bothered me about the work was their final recommendation was to use 5 times more slurry (i.e. 1 liter per minute at POU) in CMP to ' wash off ' the debris from the pad during polishing !
While both Mike and Ara make some points, I believe we are missing the big picture. This forum is not to pick apart proposals but to offer potential solutions to persistent, well recognized, technology limiting problems. Ara and Jin Goo, as well as others, have published work showing the negative impact of both polishing and conditioning debris on the process stability and wafer surface defectivity. In response to Ara's points; 1).The slurry and debris residence times are controllable using the Confluense PSM. 2). The same response applies as for item 1. The Confluense approach allows control of the slurry residence time. 3).Preliminary work was done with a TBW abrasive but there are configurations for abrasives from other manufacturers. Confluense is abrasive neutral. 4).See comment to item 3. This is a nonissue. 5). The PSM retrofit is a 1 day operation and the conditioning arm and facilities are quite simple, relative to the tool it goes on. A 3-4 month ROI seems quite reasonable and therefore not too expensive, relative to the alternatives of buying a new tool or vastly increasing the cost of consumables. Lastly, the simplest way to deal with elimination of wafer defects caused by polishing AND conditioning debris is to remove it before it can cause damage. I hope this clears up some points and gets us back on track for a meaningful discussion of proposed solutions to some serious problems that plague CMP.
As an FYI, here are some examples of pad debris as seen by confocal microscopy. Most of the data is from a collaborative work between Tohoku University (Prof. Nemoto) and Araca Incorporated, Preliminary findings indicate that pad debris generation is both a function of process (i.e. slurry type as well as tool kinematics) and the type of the conditioning disc (in this case 3M vs. MMC).
I did quite a number of presentations in attempt to attract the attention of the CMP community to the conditioning debris issue(s) since I believe 2003-4 ( "Considerations on the defect generation mechanisms in the CMP technology"). Unfortunately, being not in CMP these days, I can present my stuff only here at the CMP UG, which sessions poorly attended these days. There is a lot of very clear evidence that the Cu defectivity (scratches, chatter marks, aligned corrosion spots) are directly related to the pad conditioning debris, impregnated with abrasives and process by-products. There is one more source that people do not look at, which is laso associated with the pad conditioning unit. Although I am not sure it occurs with all the tools in the field, but with some instruments, a layer of slurry agglomerates is building-up around the conditioning disc. The thicker it becomes the more it tends to crack releasing hard chunks of it over the pad. Luckily retaining rings take care about the big chunks, however small pieces, or crashed large pieces can cause serious problems. We do (did) look what is going on with the wafer, wafer carrier and pad, and most of the problems here are already successfully solved. I believe now it is time to look more intently at the conditioning device.
Yehiel, I think the CMP community is getting your message - maybe a little late since you were certainly ahead of the curve in advocating the sources of some technology-limiting defects that plague the CMP process. Using the conditioner to spread fresh slurry and mix it with polishing and conditioning debris arguably may have made some sense when we were at the 250nm technology node but certainly not for 25nm technology! That approach to try to reach pseudo-equilibrium results in an unstable process and severe over conditioning of the pad producing more pad debris. By controlling the slurry residence time more appropriately by removing spent slurry and adding fresh, without having to mix with spent slurry, would also reduce the slurry buildup on the conditioner. Reducing the slurry flow rate would also positively impact that as well as the CoC. Does it not make sense to remove the polishing and conditioning debris from the pad and prevent their contact with the wafer as being an effective way to reduce microscratches, scratches, and chatter marks? There are other techniques proposed to reduce the impact of CMP process debris but only one that eliminates process debris from the pad.
It certainly makes sense to remove the pad debris and fragments from the pad-wafer interface during CMP so long as those fragments are not the major contributors to the CMP process and to material removal.
At this point, we have some evidence that suggests pad fragments are the PRIMARY contact areas that promote CMP, so I am not sure which way to go!
Darryl raises (in post #7) the distinction between "pseudo-equilibious" and non-equilibrious processing. Is this distinction one of the hurdles we must all cross at this point? That is, do most of us think of CMP (like any fab process) as having to be in at least pseudo-equilibrium to allow for control, or have most of us made the leap to realizing that we now must operate in inherently non-equilibious conditions?
I think most chemists (like myself) would prefer to operate near equilibrium since that is a region where one typically has the most latitude (i.e., small changes in concentrations result in small linear changes in reaction rate). Operating near equilibrium for a CMP process would potentially yield the largest process latitude, something we always strive for. The point I was making in this thread was that the pad is over conditioned due to the fact that the abrasive disk is used for a lot more than just pad conditioning. It is used to mix fresh slurry with spent slurry and debris and spread it across the pad in an attempt to stabilize the CMP process. This dilution effect practically ensures one will never reach equilibrium unless one literally floods the pad with fresh slurry and achieves pseudo-equilibrium. That greatly increases the cost of consumables and has further impact on the backend (i.e., waste disposal/treatment) costs.
Ara, I am advocating controlling the residence time of the slurry as well as the debris. If debris contributes to the removal process, it will be needed to achieve appropriate material removal rates. Data has been generated and reported that shows that higher MRRs can be achieved with lower slurry flow rates when controlling the residence time for debris and slurry on several pad types (i.e., perforated and grooved) through pad surface management. Controlling the residence time for debris allows one to lower defect generation while maintaining an appropriate removal rate to reach the optimum operating point. The side benefit is that lower slurry flow rates are possible further reducing the CoC.
Recent published studies have shown effective slurry residence time control employing advanced CMP retaining ring slurry groove designs. Increasing slurry residence time and limiting the volume of slurry also results in temperature increase in the CMP process, contributing to higher MRRs. To limit the generation of pad debris and increase the pad life it is essential to use pad conditioner in an optimum fashion, just to have enough pad replenishment and required slurry distribution. To achieve this and understand the physics of pad conditioning, it would be worthwhile to perform extensive conditioner characterization work with a number of pad conditioners (with varying degree of diamond sizes, shapes, and distribution density) and different pad designs, employing a range of contact pressures, pad/conditioner rotational speeds and slurry flow rates. To limit the cost of such characterization and have an overall feel of the relative performance of different designs, it may be desirable to perform such studies employing 2” conditioners and 6” pads on benchtop platforms first and identify the areas to concentrate in the regular size conditioner and pad studies. Some efforts in this direction are continuing and results would be shared in public domain when available. Needless to mention, even small (but consistent) reduction in slurry consumption and hence CoO, maintaining same MRR and (hopefully) improved defectivity through new CMP ring, conditioners and pads development, would be very valuable for the end users.
In No. 11 you say that '... data has been generated and reported that shows that higher MRRs can be achieved with lower slurry flow rates when controlling the residence time for debris and slurry on several pad types ...'. Do you mind sharing the data with the community?
Once I see the data, I will be in a better position to cooment.
A couple of questions: How do you plan to quantify the extent of pad debris? How would you distinguish pad debris from pad fragments or poorly supported or partially broken-off pad asperities? Each one of these play different roles in CMP.
Ara's comments (1) and (2) regarding weaknesses in the PSM seem to define a standard that is not recognized by the empirical evidence regarding slurry replacement vs. slurry dilution. At Semicon West 2009, AMAT introduced (see attached announcement) a slurry dispense point that rode on the conditioner arm, distributing the fresh slurry imperfectly but better than just using the head bow wave. For this innovation, they claimed a 15% reduction in slurry flow without loss of removal rate. The Confluense PSM takes this one significant step further and distributes fresh slurry after removing the old slurry, and any debris along with it. The notion that the concept has a weakness since the old slurry removal may be less than 100% is focusing attention on a claim that was never made. The PSM appears to remove enough slurry to justify its claim that slurry delivery has transitioned from dilution mode to replacement mode -- meaning that the wafer is responding as if it were always seeing undiluted fresh slurry. The resulting ability to maintain removal rates while reducing slurry flow extends far beyond the 15% demonstrated by AMAT. If I were a fab engineer with a cost-reduction gun to my head, I would be inclined to take a closer look at something that promises an ROI of under six months, and judge for myself whether it is too complex or too expensive.
Thanks! Yes, I do see the challenge in quantification and identification of pad debris and it has to be based on indirect measurements such as dynamic skin-friction and related tribological parameters, conditioner and pad decay magnitude and rates, MRR consistency and defectivity, etc., again not a trivial task. The pad debris from pad fragments, or poorly supported or partially broken-off pad asperities should be (relatively) minimal when the new pad is already broken-in, after initial conditioning step (e.g., detected by skin-friction achieving a nearly asymptotic value; see attached CMP-MIC paper). As a starting point, I do believe that small scale bench tests should be able to provide useful insight on pad surface condition from different duration (and operating condition) tests. This information may be used for the next level tests.
The data for slurry reduction with PSM for both perforated pads and grooved pads were presented at the NCCAVS CMPUG meeting at SCW 2009 and the entire presentation is downloadable from their website.
I was surprised that in the discussions of the impact (positive and negative) of pad and process debris, no one mentioned pad staining. Chemists know well that the build-up of end products inhibits reactions. Allowing Cu-BTA complexes to build up on the pad will have a negative impact on the MRR, dishing erosion, uniformity, and selectivity as reported by Jin Goo Park and his group. Consequently, controlling the build-up of reaction products on the pad by using a pad surface management strategy additional ability to stabilize the CMP process – what we all strive for. I have added some data from Jin Goo Park as an illustration. This was work done with normal conditioning and pad cleaning and compares various data for wafers in the lot. The variation is quite large but not unexpected.
You are right. Understanding and controlling pad staining during copper polish is very critical as well.
For those of you who may be interested in pad stain formation (and the mechanism thereof), attached is a theoretical study which just got published in JJAP.
We submitted the experimental portion of this study to TSF and it is still under review so I woun't be able to post that one for some time.
Ara, Thanks for posting your paper on pad staining. As a chemist, I tended to focus on chemical aspects in what ever I did in the semiconductor industry, even when involved with lithography or metrology. I have been surprised by the apparent lack of chemists in an industry where the manufacturing process is dominated by chemistry. I once gave a chemistry department colloquium at one of my alma maters about the role of the chemist in the semiconductor industry. I still run into chemists who graduated from that school and entered the semiconductor industry who attended that colloquium. With regard to pad staining, a benefit of PSM is to remove or minimize the build up of Cu CMP reaction products and hence minimize or avoid pad staining, yielding more consistent pad performance over the life of the pad, which would be extended relative to that without PSM. This topic may have benefitted from its own discussion thread. Maybe we can do that for the 2011 Roundtable?
I have enjoyed this discussion, as it reaffirms to me the opportunity for us all to improve this art. I would like to summarize our rational for controlling the confluense of material between the CMP work piece (wafer) and the tool (pad). The current mode of operation for rotary polishers is one that has endured decades, if not centuries – it works. Supply of the ‘lubricant’ phase of this three body system is facilitated by a pseudo-stable dilution; in the case of CMP, fresh slurry is added to a pad which is saturated with water, slurry, spent slurry, reaction products, and pad debris. The concentration of these constituents changes over time, in accordance with the dilution/replacement efficiency. We have characterized exemplary (Cu, Oxide, STI) film material responses electrochemically, and the slurry / pad, shear induced agglomeration evolution optically and ultrasonically. You would be amazed at what your wafer swims with. The part of the pad (tool) that needs to be efficiently refreshed is the tortuous surface profile and pores which hold the bulk of working fluid. Perforations and grooves contain non-working fluid ‘reservoirs’, provide little to no pressure / shear to the removal process, yet an important means to channel debris which can be entrained within. ARACA has shown that in a ‘typical’ CMP process about 90% of the slurry delivered to the pad does not participate in the polish process but goes directly to waste. Similarly in a NCAVS CMPUG presentation by Dow/R&H; Muldowney estimates 70% of the ‘steady state’ film between the wafer and pad is spent. As we scale, these inefficiencies become increasingly expensive, and we continue to battle - consumption, variability, and defect density. Controllably removing pad debris, process products, and spent slurry from the surface pores and then allowing the tool’s normal slurry delivery and carrier system to replenish slurry in the surface pores provides a means to tuning/control the residence time of the working slurry. ARACA has shown that the residence time of the saturated pad (applicable to rinse water, or slurry) can be a significant portion of the typical polish time (50-70%). This leads to an inherent reaction decay mechanism as the chemical character of the fluid between the pad and wafer is unstable, and if qualitatively compared to virgin material, less efficacious. Switching from a mechanical mixing / dilution dominated approach to evacuation and replenishment mode shows rate stability at a broader flow range (links to data have been posted showing this effect in a previous entry) This working film state – chemical efficaciousness, slurry utilization (fresh/spent) efficiency, effective abrasive contact area (rolling LP pad asperities, fixed textural asperities), and effective abrasive morphology,(slurry solids, gel’s, other) is what our enhancement provides users a means to control – regardless of process materials, (pretty effin exciting!).Lastly, removing or controlling the residence time of pad debris may have a significant impact on LPDs and microscratches, as pointed out by Yehiel. Scratch defects can have a significant impact on yield and reliability. I was recently in a customer’s site where they were running 6-inch wafers with 220nm technology and their principal problem was periodic yield and reliability issues due to microscratches. I have attached three slides comprised of data from Berkeley, Rohm and Haas, and ARACA that support the above data for slurry waste and the impact of managing the residence time of the working fluid. Thank you for the forum, and the important work you all contribute to this art.
This has been a very fruitful discussion topic, and (as Darryl suggests) we probably have at least one or two related ideas to discuss in the next virtual roundtable.
The initial question did not explicitly limit the consideration of pad debris to physical "chunks" of matter (as opposed to chemical "residues" of matter), but that was certainly most of our discussion here. So, IMHO we should definitely re-visit the issues of pad staining in our next roundtable, perhaps along with consideration of initial skin, seasoning, and conditioning.
I really enjoyed contributing to this discussion topic and I learned quite a lot from others as to the issues and nuances surrounding pad debris.
I currently have one PhD student whose thesis is primarily focusing on pad debris generation and quantification in CMP, so, in the future, I should have lots more to contribute by way of results.
In the meantime, we need to ASAP determine how pads from CMC (D100), JSR and Dow (IC and Vision) differ in terms of the debris that they generate (both with a common diamond disc as well as with their own recommended BKM discs). I think the inherent differences among these pads will certainly shed light on their potential contribution to (or avoidance of) defect generation.
Some of the pad makers will not volunteer such information since the results may be potentially adverse due to inherent issues with their material of construction, porosity and weak pore sidewall structure. Other, more open and progressive pad makers, may provide such information in order to better position their products and to gain market share.
Whatever the situation with the suppliers, academia should take the lead on getting such informatin out in the public to the extent possible.
Overall very interesting and useful discussions around the topic of pad debris. Lots of aspect of problems associated with pad debris and potential solutions have been discussed and there is no need to repeat what has already been said. But one aspect that was discussed in details is role of mechanical properties of pad asperities and role it may play in amount, shape, size of the pad debris during polish. There have been attempts to better characterize pad asperity mechanical properties by both academia (MIT, UofA, Uof Iowa, etc,) as well as pad suppliers but I believe more work in this area is still needed and should be combined by better understanding of chemical (surface) properties of pad asperities. One may argue that pad debris could be bigger chunks and is not necessarily limited to topmost surface (i.e pad asperities) and that could be true. Nevertheless, deeper understanding and better characterization of pad asperities can contribute to better management of pad debris and limiting its negative influence on defectivity.
I, too, am pleased to see that this topic generated so much productive discussion. It's not often that a problem of this nature has a potential solution available (Confluense PSM) for users to evaluate and actually begin to answer some of these questions.I expect to have a lot more published data available to us all by this time next year.
Mansour, I am pretty sure that the academic work to which you refer in #24 is taking place at my alma mater, Iowa State in Ames, and not U of Iowa in Iowa City. It's a common mistake, but not one that will allow me to sit idly by.
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 (certainly this one) 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
