Future Cu CMP over ultra-low-k dielectrics (ULK) may
require almost zero downforce; will we finally have to use electro-CMP (ECMP)
or can chemistry alone save us?
I think ECMP is pretty much dead. End-users and tool makers have given up on it. Even academicians have dropped ECMP from their long term research roadmaps.
Regarding ULK dielectric delamination during CMP, I think three main factors are at work: (1) average value of the shear force, (2) fluctuation of the shear force, and (3) polish time. Practically, all three responses need to be reduced in order for us to realize high yields in HVM.
This way of thinking is not much different from formally assessing the structural damage following seismic events where the mean magnitude of the earthquake, as well as its fluctuating component, and the total duration all are known to contribute to the overall damage.
Regarding the reduction of mean shear force during polishing, in some of the published literature, I have noticed that people tend to confuse the term ‘low shear’ with ‘low pressure’ (they tend to use them interchangeably). In reality shear force is the product of coefficient of friction (COF) and pressure. As such, there are two ways to reduce shear force. The first and obvious way is to lower the wafer pressure through mechanical design and engineering. Modern polishers now can controllably achieve wafer pressures of less than 0.5 PSI which, even with relatively high values of COF (i.e. 0.5 for copper or Ta polishing), impart low shear forces on the wafer. The other approach is to reduce COF through several means such as changing the pad groove design, using conditioners that impart a certain micro-texture on the pad surface that promotes low COF, changing the kinematics of the polisher (i.e. high rotation rate of the head) or using additives and precision engineered nano-particles in the slurry. Any of these factors alone can easily reduce COF by a factor of 2.
Regarding the reduction of shear force fluctuation, pad groove design, optimization of polisher kinematics and pad micro-texture engineering are the main factors as any one of these factors, if properly optimized, can reduce shear force fluctuation by 2X.
Regarding reduction of polish time, the ITRS is on our side (for a change) since films are becoming thinner and thinner and as such polish times are getting shorter and shorter. To further reduce the polish time (without having to increase pressure) one can slightly increase the platen temperature through controlled platen heating during CMP. For copper CMP, assuming an apparent activation energy of 1.0 eV corresponding to the chemical formation of the copper passivation layer, a 5 degree Celsius increase in the temperature can increase the rate by as much as 50 percent thereby reducing polish time by a factor of 2.
eCMP could have been a good solution for low down force processing. eCMP. In concept eCMP is capable of providing a anisotropic removal component were "E" would have replaced "M". In practice however, "E" seemed to replace "C".
In terms of limiting damage, low ultimate shear force or maximum shear force would need to be below the mechanical limits of initiating a failure mode - cohesive (material) or adhesive (interface). I would add that keeping both normal down force and shear force low is desirable, since while a 50% lower COF pad may give half the shear force for a given down force, it could still initiate damage due to high normal force. This inevitable leads to removal rate mismatches for different slurries, since slurries are designed around standard, "normal COF" pads. The COF vs Removal Rate behavior shows significant differences between slurries.
Ideally slurry chemistry could be tuned to provide removal initiation at low net shear force for optimal pad-slurry combination to enable ultimate low shear force with low normal force process at high thru-put.
Modulating temperature to achieve higher removal, as suggested by Ara, is really a proxy for tuning chemistry.
I recall as fascinating presentation by Paul Fisher of Intel at Lake Placid CAMP three years ago, in which he demonstrated that the mechanical strength of ULK dielectrics was much greater than the forces generated in CMP. Sorry I am unable to find the data-- maybe Monsour can dredge it from his files. But up to now, we haven't seen damage as an issue. The use of a TEOS-CVD cap over ULK has been effective, and we have developed barrier slurries at Cabot Microelectronics that polish through the TEOS layer and stop on ULK with no reduction in Keff. So this is a long-winded way of saying that I don't believe eCMP will be needed.
Sorry Cliff, I don't have Paul's presentation handy but I'll dig through my files again. In general, I tend to agree w/ the comments/observations by Cliff about ULK mechanical damage during conventional CMP not being a major issue --at least so far-- and hence eCMP may not be necessary. However, shift in K as a result of penetration of chemicals into porous low K film during direct ULK polish is well documented and published and needs to be taken into consideration when desiging the polish process for Cu/low K stack.
Just a quick comment on Monsour's observation about Keff drift. I think this is a significant issue for higher pH barrier slurries-- one reason we have lowered our latest product to 4-5. We are giving a paper at the CSTIC-ECS conference in Shanghai about it which should be published in the conference book and hopefully in a special ECS issue.
Ara, just a quick comment on COF. The COF is just a formal parameter linking the normal force and the friction force, and it is function of so many variables, especially in the CMP case: lubrication (slurry composition), temperature, abrasive content (slurry+ debris), wafer surface (pattern density, average and local), lubricant supply over the interaction area, pad wear profile, relative velocity ( hydroplanning effects), and many more. To my opinion, one of the major factors is pad surface state. Wafer never sees the actual bulk pad material. What is engages with is a surface which is strongly modified by the conditioning+process interactions. Surface roughness is created by the pad conditioning and strongly depends on its design and parameters. Pad asperities - these are the objects wafer actually engages and their shape and size are ultimately defining the interaction, and friction, and the planarization efficiency and other CMP performance parameters. And this is the conditioner who creates the asperities, so as the conditioning so the CMP performance. This is my solid opinion and this is why I believe the conditioning design/parameters are the key matters defining the COF, and ultimately the overall planarization performance.
As far as the ECMP (or whatever planarization approach involving electrochemistry) is concerned, here is my couple of cents (remember I was teaching Physical Chemistry for more about two dozen of years, and do know what I am talking about): For the spontaneous chemical processes (and Cu oxidational removal is such a process) there no stuff that electrochemistry can do but conventional chemistry cannot. In CMP, as you definitely know, the chemistry modifies the material making it more pliable for removal, and tangential mechanical action scrubs it away providing planarization effect. Delicate balance of these two actions allows CMP to perform "as doctor prescribe". This is why there is no other way to perform high quality planarization, and this is why CMP people have good job security. ECMP does not offer anything beneficial just unnecessary and serious complications new- this is (actually was since ECMP just appeared at the horizon; Sass is a witness) my opinion, so I I was not surprised at all when the tools were shipped back to the vendor. What I was surprised with was that somebody was that short-sighed and paid $$ for the ECMP business- that was a big surprise for me.
Anyway, I do not see any even niche application for ECM, unless it does not require planarization efficiency. Than it should be re-named, like, say, electro-etching, electro-erosion, etc.
Thx for your comments (Reply No. 6). I totally agree with you regarding pad surface micro-texture as being the most important parameter in CMP process performance. As such, the role that the conditioner plays is paramount.
Ed, today was my day off (formally) so, I guess, I made my contribution to the discussions. I doubt I will be able to do too much in the incoming days. So thank you so much for inviting me in (well, for some reason, my name had CMP-ed out from the list).
I wonder if Prof. Gotkis didn't overstate his argument that eCMP offers nothing that can't be accomplished with chemistry alone. Certainly that is true thermodynamically (a word which I will admit to having used rarely since graduating in the dark days) but the use of applied fields does a lot to change the kinetics. Also, shifting the applied voltage can make the surface materials react in an environment in which they would otherwise be chemically inert, or at least incredibly stubborn to remove. I agree that eCMP adds a lot of complexity, but I think it still has a chance to be called upon to solve problems that aren't being resolved any other way. I learned last year that one of the hard disk manufacturers was experimenting with eCMP, but they never followed up to my inquiries for more details and results. So this lead is either dead or so successful that it is now an incredibly valuable trade secret. Any bets?
Finally...ECMP for HDD...per my experience, ECMP for Nickel Phosphorus based HDD substrates should be able to adapt into 1 step and completely replace alumina (1st step) and silica (2nd step) slurries yielding better surface quality...but why bother when 63.5 inch glass substrates are kicking out 95mm NiP/Al substrates from the future market.
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.
I think ECMP is pretty much dead. End-users and tool makers have given up on it. Even academicians have dropped ECMP from their long term research roadmaps.
Regarding ULK dielectric delamination during CMP, I think three main factors are at work: (1) average value of the shear force, (2) fluctuation of the shear force, and (3) polish time. Practically, all three responses need to be reduced in order for us to realize high yields in HVM.
This way of thinking is not much different from formally assessing the structural damage following seismic events where the mean magnitude of the earthquake, as well as its fluctuating component, and the total duration all are known to contribute to the overall damage.
Regarding the reduction of mean shear force during polishing, in some of the published literature, I have noticed that people tend to confuse the term ‘low shear’ with ‘low pressure’ (they tend to use them interchangeably). In reality shear force is the product of coefficient of friction (COF) and pressure. As such, there are two ways to reduce shear force. The first and obvious way is to lower the wafer pressure through mechanical design and engineering. Modern polishers now can controllably achieve wafer pressures of less than 0.5 PSI which, even with relatively high values of COF (i.e. 0.5 for copper or Ta polishing), impart low shear forces on the wafer. The other approach is to reduce COF through several means such as changing the pad groove design, using conditioners that impart a certain micro-texture on the pad surface that promotes low COF, changing the kinematics of the polisher (i.e. high rotation rate of the head) or using additives and precision engineered nano-particles in the slurry. Any of these factors alone can easily reduce COF by a factor of 2.
Regarding the reduction of shear force fluctuation, pad groove design, optimization of polisher kinematics and pad micro-texture engineering are the main factors as any one of these factors, if properly optimized, can reduce shear force fluctuation by 2X.
Regarding reduction of polish time, the ITRS is on our side (for a change) since films are becoming thinner and thinner and as such polish times are getting shorter and shorter. To further reduce the polish time (without having to increase pressure) one can slightly increase the platen temperature through controlled platen heating during CMP. For copper CMP, assuming an apparent activation energy of 1.0 eV corresponding to the chemical formation of the copper passivation layer, a 5 degree Celsius increase in the temperature can increase the rate by as much as 50 percent thereby reducing polish time by a factor of 2.eCMP could have been a good solution for low down force processing. eCMP. In concept eCMP is capable of providing a anisotropic removal component were "E" would have replaced "M". In practice however, "E" seemed to replace "C".
In terms of limiting damage, low ultimate shear force or maximum shear force would need to be below the mechanical limits of initiating a failure mode - cohesive (material) or adhesive (interface). I would add that keeping both normal down force and shear force low is desirable, since while a 50% lower COF pad may give half the shear force for a given down force, it could still initiate damage due to high normal force. This inevitable leads to removal rate mismatches for different slurries, since slurries are designed around standard, "normal COF" pads. The COF vs Removal Rate behavior shows significant differences between slurries.
Ideally slurry chemistry could be tuned to provide removal initiation at low net shear force for optimal pad-slurry combination to enable ultimate low shear force with low normal force process at high thru-put.
Modulating temperature to achieve higher removal, as suggested by Ara, is really a proxy for tuning chemistry.
I recall as fascinating presentation by Paul Fisher of Intel at Lake Placid CAMP three years ago, in which he demonstrated that the mechanical strength of ULK dielectrics was much greater than the forces generated in CMP. Sorry I am unable to find the data-- maybe Monsour can dredge it from his files. But up to now, we haven't seen damage as an issue. The use of a TEOS-CVD cap over ULK has been effective, and we have developed barrier slurries at Cabot Microelectronics that polish through the TEOS layer and stop on ULK with no reduction in Keff. So this is a long-winded way of saying that I don't believe eCMP will be needed.
Ara, just a quick comment on COF. The COF is just a formal parameter linking the normal force and the friction force, and it is function of so many variables, especially in the CMP case: lubrication (slurry composition), temperature, abrasive content (slurry+ debris), wafer surface (pattern density, average and local), lubricant supply over the interaction area, pad wear profile, relative velocity ( hydroplanning effects), and many more. To my opinion, one of the major factors is pad surface state. Wafer never sees the actual bulk pad material. What is engages with is a surface which is strongly modified by the conditioning+process interactions. Surface roughness is created by the pad conditioning and strongly depends on its design and parameters. Pad asperities - these are the objects wafer actually engages and their shape and size are ultimately defining the interaction, and friction, and the planarization efficiency and other CMP performance parameters. And this is the conditioner who creates the asperities, so as the conditioning so the CMP performance. This is my solid opinion and this is why I believe the conditioning design/parameters are the key matters defining the COF, and ultimately the overall planarization performance.
Best,
Yehiel
As far as the ECMP (or whatever planarization approach involving electrochemistry) is concerned, here is my couple of cents (remember I was teaching Physical Chemistry for more about two dozen of years, and do know what I am talking about): For the spontaneous chemical processes (and Cu oxidational removal is such a process) there no stuff that electrochemistry can do but conventional chemistry cannot. In CMP, as you definitely know, the chemistry modifies the material making it more pliable for removal, and tangential mechanical action scrubs it away providing planarization effect. Delicate balance of these two actions allows CMP to perform "as doctor prescribe". This is why there is no other way to perform high quality planarization, and this is why CMP people have good job security. ECMP does not offer anything beneficial just unnecessary and serious complications new- this is (actually was since ECMP just appeared at the horizon; Sass is a witness) my opinion, so I I was not surprised at all when the tools were shipped back to the vendor. What I was surprised with was that somebody was that short-sighed and paid $$ for the ECMP business- that was a big surprise for me.
Anyway, I do not see any even niche application for ECM, unless it does not require planarization efficiency. Than it should be re-named, like, say, electro-etching, electro-erosion, etc.
Thanks,
Yehiel
Hi Yeheil:
Thx for your comments (Reply No. 6). I totally agree with you regarding pad surface micro-texture as being the most important parameter in CMP process performance. As such, the role that the conditioner plays is paramount.
I am glad that you highlighted this matter - Ara
Ed, today was my day off (formally) so, I guess, I made my contribution to the discussions. I doubt I will be able to do too much in the incoming days. So thank you so much for inviting me in (well, for some reason, my name had CMP-ed out from the list).
Best,
Yehiel
Regarding #7
I wonder if Prof. Gotkis didn't overstate his argument that eCMP offers nothing that can't be accomplished with chemistry alone. Certainly that is true thermodynamically (a word which I will admit to having used rarely since graduating in the dark days) but the use of applied fields does a lot to change the kinetics. Also, shifting the applied voltage can make the surface materials react in an environment in which they would otherwise be chemically inert, or at least incredibly stubborn to remove. I agree that eCMP adds a lot of complexity, but I think it still has a chance to be called upon to solve problems that aren't being resolved any other way. I learned last year that one of the hard disk manufacturers was experimenting with eCMP, but they never followed up to my inquiries for more details and results. So this lead is either dead or so successful that it is now an incredibly valuable trade secret. Any bets?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.
Happy planarizing.