Advanced Metallization Conference 2021
30thAnniversary, Asian Session

  

Publication of ADMETA commemorative booklet:
   Chapter 2:
 From round-table discussion in 2007

     The rise of multilevel interconnects technology development

     Previous Page                      Back to contents

  Planarization, selective W-CVD, and Cu damascene

Masahiro Kashiwagi, Takahiko Moriya, Takayuki Ohba, Nobuyoshi Kobayashi, Observers: Editorial Board Members (Shingubara, Tsujimura, Masu, Kondo)

1. Establishment of ADMETA

Shingubara (Moderator, former Toshiba): A brief report from the first meeting in 1989 listed Mr. Kashiwagi's name. He is vice-chairman. At first, you were doing it in the U.S. Tungsten workshop.
Moriya (former Toshiba): That started in 1984. I went there for the second time in 1985. Blewer (R.S. Brewer, Sandia National Lab.) started caring about Tungsten in Japan.
Moriya: Were you present in 1986?
Ohba (former Fujitsu): Yes, I think it was the Xerox Center, where I met Mr. Kobayashi.
Kobayashi (former Hitachi): Yes, I am out, too. Thus, for the first time, I think Mr. Ohba and I met.
Kashiwagi: When did you name ADMETA?
Shingubara: ADMETA, I am sure Ohba-san did it. That is, the tungsten workshop was turned into the Advanced Metallization Conference. This has occurred since 1991. First, it was called as “workshop on tungsten and other refractory metals.”
Shingubara: Who was the first Japanese to attend the workshop?
Ohba: The first time, I am not sure anyone attended it.
Shingubara: Therefore, the day's ADMETA event started in 1988 when you talked to Blewer in Albuquerque.
Kashiwagi: Blewer wrote to me, what should I do? Therefore, I think Prof. Furukawa -- Seijiro Furukawa -- Tokyo Tech was the best because he was doing a three-dimensional integrated circuit, thus I thought he was the best. In those days, if you had a problem, there was always Prof. Furukawa.
Ohba: At that time, there was a project somehow for the next generation... (Note: Next Generation Industrial Infrastructure Technology Research and Development for 3D circuit supported by former MITI, 1981-1990)
Kashiwagi: It is the New Function Device Research and Development Association. (Note: Academic Society Secretary). The whole of this party was there. All the three-dimensional ones were hanging out, drinking all year round, and discussing various things. After all, Prof. Furukawa was the best leading person for this conference activity in Japan. Because interconnect will be quite important in the future, we talked about organizing a substantial conference that would not cost a lot of money; thus, we held it at Tokyo Tech.
Masu (Tokyo Tech): Centennial Hall (Museum)
Shingubara: I think it was just brand-new.
Masu: That is right. It was brand-new in November 1989. (Note: opening of the new hall, November 1987)
Kashiwagi: I thought it was a little-bit lonely place; however, I remember that it was appropriate for a research meeting, and we did it.
Ohba: It was 130 participants.
Shingubara: This was around the time of the bubble boom. The company had a good economy.
Kobayashi: Japan was in excellent condition at that time.
Shingubara: Japan as No.1. If you look at the Red Book at that time, there were some pictures of Brewer, Mr. Kashiwagi, Prof. Furukawa, etc. However, their quality is poor.
Ohba: At that time, I pasted the photo and graph on a manuscript prepared by a word processor.
Kashiwagi: I did it too. In those days, computers were used only for word processing. (laughs).
Shingubara: The word processing was not yet widespread, right?
Ohba: I used an IBM typewriter.
Shingubara: It is recorded that communication was done by fax. Handwritten fax. It is a fax of typed English to a foreign country.
Ohba: It was a a Telex (Telegraph-exchange) era.
Kashiwagi: Times have changed a lot.


2. The Beginning of Multilayer Interconnects and the Impact of Selective W-CVD

Shingubara: I would like to go over the flow of interconnect technology, looking back at the past and asking various questions to those who understood. What led to the creation of the tungsten workshop?
This is a simple question; however, I actually joined Toshiba in 1985, and these two people (Mr. Kashiwagi and Mr. Moriya) took very good care of me and trained me. At that time, we used Al two-layer interconnect, or at most three layers. When we talk of multilayer interconnect, we started with Al two-layer it was it at the early stages?
Kashiwagi: Al bilayer has been used for a long time.
Moriya: It was used for Bipolar semiconductors.
Shingubara: Has this been used since the early 1970s?
Kashiwagi: Multilayer Al interconnect has been required since the early days of LSI. Everyone tried it but it did not work very well because of the problem of step coverage. We could not perform the LSI microscopic ones; however, we used some of the bipolar ones, which were quite loose.
Shingubara: Did you start using Al bilayer from bipolar?
Kashiwagi: With bipolar, multilayer interconnections were needed in various places, thus, in reality, we first tried it. However, there were the so-called step coverage problem and the problem of step breakage.
Moriya: I think it was just before the 80s. Of course, two-layer interconnects existed in the bipolar era; however, when logic devices became widespread, such as 6k gate arrays, it was decided that two-layer interconnects must be used. This was when we moved from the 3 µm to the 2 µm rule. Wet processing was not appropriate for the movement. That is why we had to go with the RIE. I think it was just in the 1980s.
Kobayashi: That is, where I was. I remember well because I joined the company in 1980. At that time, the DRAM capacity was still 1M byte, and it used an Al single layer. I worked very hard to introduce the metal gate to reduce the gate resistance.
Shingubara: Was it 1M?
Kobayashi: Yes, it was developed as the most advanced DRAM. At that time, Al remained in a single layer.
Shingubara: Was it still a single layer at that time?
Kobayashi: Then, there was a discussion about lowering the gate resistance instead of the Al interconnect resistance. In the end, Al had successfully become a double layer, and metal gates were delayed to be introduced until recently where they were necessary for practical applications. Thus, at that time, dual-layer interconnections were still quite an amazing technology.
Kashiwagi: There was almost no such thing as two-layer interconnects.
Shingubara: Multilayer interconnects technology started in the 1980s. This was what I wanted to hear.
Moriya: At that time, CVD was used for insulating films, and sputtering was used for metals. This was the usual way of living; however, sputtering may have reached its limit in terms of coverage, thus metal CVD came into focus. At that time, in the 1980s, W started to be employed in various applications.
Ohba: I would like to ask why Toshiba used Molybdenum in those days. We used Tungsten Silicide (polycide), and because there was a Tungsten CVD process. It was the way of Fujitsu.
Kobayashi: In those days, there was much discussion about whether to use Mo or W materials. This is a long-standing debate. Each company had its own likes and dislikes.
Kashiwagi: There was a time when everything was done, and there was an example of Mo silicide made by CVD. Finally, MoF6 was more difficult to use than WF6 because of its activity, thus W was employed, including sputtering. What was most shocking was that selective-W became increasingly popular, and Blewer was doing its best to promote it.
Kobayashi: Did W-CVD start around 1980?
Moriya: This was done in the 1970s. I did not intend to select. Let us try W first.
Ohba: You used horizontal furnace-type equipment, didn't you?
Moriya: I tried adding W; however, the results showed that it did not stick to the oxide film.
Kobayashi: I think the U.S. published epoch-making technical data, and Blewer was working on it,
thus we decided to gather such technical colleagues regarding W material. Moriya: I thought it was the ideal way to fill.
Shingubara: Indeed, it is. Was 1980 the exact year of selection of W?
Moriya: Of course, it existed before that; however, it started being employed at that time.


3. Early Planarization

Kashiwagi: Initially, wet etching was used in two-layer interconnects. To prevent step breakage, we attempted to make it as not bumpy as possible. We also performed anodic oxidation of Al. We turned ordinary Al into alumina and made multilayer interconnects. Therefore, the bumpiness is reduced to a minimum. This approach also existed.
Shingubara: Were you very serious about anodic oxidation?
Kashiwagi: Yes, we did. I worked on it quite a bit.
Shingubara: I have read a paper written by someone from IBM.
Kashiwagi: When anodizing, some parts of the Al become alumina and some remain metal, which is a good reason. However, there is the same issue with the wet etching. It is not so appropriate for miniaturization because of side etched.
It was at the inception of gate arrays that multilayer interconnects were seriously miniaturized in LSI. Logic has been expecting it for some time. Memory was not so. Everyone in the industry was concerned about the bumpy problem of step coverage. Everyone in the industry was irritated. Before that, phosphosilicate glass reflow was used; however, the Al process could not be used because of its low temperature. Subsequently, Mr. Honma (Yoshio Honma, Hitachi, Ltd.) said, "I am going to do this, flattening. I told Ogawa-san (Shinichi Ogawa, Matsushita) that the other day, and he said, "I was the first to name it flattening.
Shingubara: Is that so?
Kashiwagi: At least in Japanese. Honma-kun was the first to challenge flattening. This is still in the ECS paper. He applied the resist and made it wet. And, etchbacked it. It is the first time in Japan.
Kobayashi: That was the first time, wasn't it? It was very tricky. You put the resist back on!
Kashiwagi: At the time, I said a lot of things like, "What is the point of putting it back if it is already attached?” Was Moriya-san working on the etch-back of nitride at the time?
Moriya: Yes. There were various ways to use the etch-back at that time.
Kashiwagi: I found that the nitride was quite successful. Toshiba decided to follow that line. Honma-san published a paper on it, and I thought it was more clever (laughs).
Ohba: It was. There were two approaches. I mean it was self-planarization of dielectrics using spin-coating and mechanical polish. There is different technology direction even in the same company.
Kashiwagi: Within Toshiba, there was a lot of controversy over such flattening.
Ohba: I think the etching group was probably strong among Toshiba R&D.
Moriya: In those days, I felt that the processor was stronger and pushed through.
Ohba: Yes, an opinion from process engineers was strong enough to decide technology selection. In short, it means there was still enough margin for the process, for example, either choosing spin-on or etch-back.
Kashiwagi: It was.
Ohba: In the 1980s, there was a margin to optimize in the interconnects process even when changing one process technology. However, the process margin among processes became severe following generation change. This means that there needs high accuracy and precision not only process but also process equipment. It became impossible to use a new single process for an advanced generation. Thus, it was becoming increasingly important to develop a combination of process modules accompanied by equipment.
Kashiwagi: That is true.
Ohba: How about in the case of Hitachi? Did the circuit design team as strong drive development?
Kashiwagi: The design was reliable.
Kobayashi: The design group was not so much reliable. However, it was the profit center and controlled the profit in Hitachi.
Ohba: They could act big attitude. How was it Toshiba?
Kashiwagi: This is because of the grudge. There was a time when I was a material processor, and at Toshiba, elites were always device manufacturers. There was a time when elites were always device engineers, and we were fiercely opposed to that. We created an era in which process engineers would take the lead.
Ohba: In my experience, I was always in the supercomputer competition how to win the race. So, I may just develop a process that can take as number one in the world.
Kashiwagi: Is Fujitsu is working on new technologies, in this regard?
Ohba: Designers would say, “Bring us whatever interconnect technology as long as it can use at the high current density.“ It was an amusing day.


4. Metal gate: Lessons learned from MoSi2

Kashiwagi: I wonder who had the most metal gates in the world in the past. It is GE, I think.
Kobayashi: Yes, it is GE, and Shinha (A. K. Shinha) of Bell Labs who have started the metal gate as pioneers.
Kashiwagi: Shinha used to develop metal gates. That is, in 1975, I think. Mochizuki-san came when I was working, and since the silicon gate had just ended, we decided what to do next. The next step was a metal gate.
At that time, we were all making our own equipment. The heater material was a Mo silicide called Super Kanthal. This was made by Toshiba Ceramics. I asked them to provide some pieces of it, so I propose to mochizuki-san to spatter on it. When putting it on, I found that it glittered and had a very low resistance. I thought that this might have worked. Thus, my biggest desire was to make it compatible with the Si gate process. Because Mo silicide is often used as a heater, it fails to oxidize. Therefore, I thought that this could be quite useful, thus I did a lot of work, transistor characterized, and studied the compatibility of the Si gate process. Some of the device engineers at Toshiba thought that because the specific resistance was one digit lower than that of the Si gate, it was quite possible. They said something about submitting it to an academic conference. The context of the two is somewhat ambiguous; nevertheless, the Mo gate came first. Therefore, we presented it in the ECS. It was quite popular at that time. No one else in the world was doing it yet. To the gate. However, at that time, Sinha was working on W gates in the Bell Labs. He also had issues with the W oxidation problem.
Shingubara: Is that still sputtered W?
Kobayashi: Bell Lab initially used an EB evaporation. Sputtering began in the 1980s and became mainstream for the deposition method of refractory metals.
Kashiwagi: In those days, if we went on a business trip overseas, it was usually to IBM in Yorktown Heights and Bell Labs in Murray Hill. He was a W, but he said that silicide was interesting. When I went to IBM, they had already started considering silicides, and, of course, they had not yet made LSI. At that time, they were trying to decide whether W silicide or Ta silicide was better.
Kobayashi: In the 80s?
Kashiwagi: No, it was earlier, in 1977 or 1978.
Kobayashi: Probably the original work was done in the 1970s. Gradually, it has attracted more attention than the metal gate. I think it was in 1982 or 1983.
Kashiwagi: Subsequently, Murarka (S. P. Murarka, Polytech Inst.) performed a lot of work.
Kobayashi: Yes. At that time, there was still a debate about whether to use W, Mo, silicide, Ta, or Ti silicide.
Kashiwagi: At that time, Ti silicide was not used as a gate. We never thought of it.
Kobayashi: Ti silicide was not used only after Salicide was introduced?
Kashiwagi: It was after the release of Salicide.
When I talked to IBM, they said, "Oh, I see, Mo (Silicide). Then they came back and said, " we are starting thinking of it at IBM. Thus I said, "I will have to perform LSI first. Toshiba chose Mo silicides, and they were much wailing. Toshiba settled for the Mo silicide, and they were going crazy. However, as I worked on it, I gradually came to realize that the process of the W silicide was better than Mo.
Saraswat (K. C. Saraswat, Stanford University) was not doing anything at that time; however, he said silicide was interesting. He immediately began to investigate W silicide and silicide. Subsequently, Ti silicide appeared in a flash. Therefore, my lesson is that even if the first time goes well, you have to compare it calmly.
Finally, in multi-layered interconnects, the steps in the interconnects can be flattened; nevertheless, the contacts and holes are not. Therefore, I thought W-CVD will work to embed them? I thought, such that we can make complete planar metallization? So, Moriya-san first reported on embedding at IEDM.
Shingubara: That was at IEDM in 1983.
Kashiwagi: I think we made two or three layers of planar metallization. I think this was the first time we embedded it.
In the Toshiba story I mentioned earlier, I said that Mo-silicide was good; however W also said that selection was good, thus I specialized in that. The final answer was slightly different.
Ohba: It was a good day. My boss asked me, “Why is it okay for Toshiba to use Mo and W?” Then, I had to make a counter-proposal that NOT needs to evaluate Mo to overcome Toshiba (laughs).
Moriya: There was W at that time; however, I think Mr. Mogami (Toru Mogami, NEC), who is now at Selete, filled Mo with bias sputter. I think it was in 1982 or later. He neatly filled 1 µm or 2 µm holes.
Kashiwagi: That is right. Was there a bias spatter?
Kobayashi: The quality of the film was not so good because it involved Ar. Finally, bias sputtering was not very effective; however, there were several candidates for contact and via filling, including selective plugs of W, bias sputtering of Mo silicide, and bias sputtering of Al at that time. That was in1985.
Kashiwagi: Bias sputtering subsequently disappeared.
Kobayashi: It disappeared for contact and via filing, but it was still being developed for another approach at that time. Mr. Homma likes the planarization of interconnects. Thus, he was always sticking to it, and his vision in the future would be based on the etch-back of insulating films and bias-sputtering of Al. Incidentally, I started my carrier related to devices. Because I started from metal gates and wondered which was better, W or Mo. Toshiba said the last is W (silicide). NTT is also Mo.
Kashiwagi: NTT has been doing a good job with metal gates for a long time. I have heard of the resistance of really thin metals. They have been performing very well at basic levels.


5. We made the equipment ourselves.

Kobayashi: I think the situation changed dramatically when CVD became overwhelmingly easy to handle for W materials. I think it could also have been used for Mo silicide. There was a lot of discussion at the time. However, it is difficult to develop other metals by using CVD.
Kashiwagi: Silicide was mostly sputtered at Toshiba. Okumura (Katsuya Okumura, Toshiba) did a lot for us. Finally, there was stress on the downside, and I had to cut it off. I started using W in combination at that time, and I thought that W had a wider process margin, thus I moved on.
Kobayashi: This is probably because Toshiba was quite advanced in silicide. First, it was sputtering, right? However, Hitachi was a little behind, thus we ended up W silicide CVD with Genus tool for mass production. The problem of interconnecting cut-off also occurred with silicide due to poor step coverage, thus I thought that CVD would be more advantageous than sputtering.
Kashiwagi: As far as film formation is concerned.
Kobayashi: That is why we ended up with W silicide, including mass production. It is not just a matter of quickly getting it off.
Kashiwagi: This is not what I meant. That was a lesson learned, and if you chose the right one. Although Selete was performing its best, the second-best from Taiwan was watching closely.
Kobayashi: Is that what mass production technology entails? I am not blunt when I say this.
Kashiwagi: In those days, we enjoyed technology. Moriya-san would have chosen W.
Ohba: Is it Okay that I may understand that Toshiba had many engineers push ahead sputtering technology, whereas CVD technology in Fujitsu?
Kashiwagi: Ever since Maeda-san (Kazuo Maeda), CVD has been Fujitsu.
Ohba: Fujitsu has a long history of CVD. So, there was no argument when we look CVD, everyone is OK with it.
Kashiwagi: Therefore, it is not a matter of technological inevitability but rather by chance.
Ohba: How about Hitachi on it?.
Kobayashi: No, sputtering was in the main. However, the refractory materials I mentioned have increasingly gone to CVD. Al-based products are overwhelmingly sputtered, and there is no doubt that sputtering technology is the basis of the interconnect process. However, CVD is the dominant technology for the contact area and gate. As I said, bias sputtering is not the best for contact filling.
Tsujimura (EBARA): What is interesting is that each company has a reliable proponent. Is it Mr. Maeda for Fujitsu, Mr. Okumura for Toshiba, and Mr. Homma for Hitachi?
Kobayashi: Mr. Homma, yes.
Tsujimura: In the 1970s, when you discussed whether to use CVD or PVD, did each device manufacturer have its own equipment technologies? In the 1980s, however, equipment manufacturers started being independent.
Kobayashi: Was it a good time for people to design their own processes?
Tsujimura: This was truly a time of learning from the past. As someone like me who came in later, I am reminded that technology repeats itself. Nitride and resist are deposited and then planarized by etch-back, was RIE replaced by CMP? This is the same as planarization using removable technology.
Was the idea of planarization mentioned? For example, TEOS-03 or SOG.
Kobayashi: What we are doing may be much more complicated than in the past.
Tsujimura: What was done in the 70s was also done in the 80s, 90s, and even today. I hope that this booklet will be used to learn something new from history.
Kobayashi: Therefore we do not learn history in the past?
Shingubara: Even if you are the first to work on it, you cannot know the rest.
Kashiwagi: It is impossible to know everything. It will be sick of knowing. Everyone must go through the same thing.
Ohba: Finally, that is “you never know unless you try.”


6. The Night Before Cu Damascene

Kashiwagi: In that context, I am still curious about Fujitsu's product that added Ti to Cu to form TiN on the surface.
Ohba: That disappeared once.
Kashiwagi: Did it appeared and then disappeared again?
Ohba: Again appeared. However, the principle is same as the Fujitsu did, which means the use of out-diffusion behavior for the Cu compound.
Kashiwagi: I wondered if everyone knew exactly what that job was as if it had been there for a long time.
Ohba: No, I do not.
Kashiwagi: It is chic, and they call it a new principle and publish it in the newspaper. However, was Fujitsu not doing the same thing?
Kobayashi: In semiconductor technology, do we have a completely new concept?
Kashiwagi: Hardly at all (laughs).
Kobayashi: I think it is no problem. It is a production technology we are developing. Thus, I think we should apply some part of the old technology and if it is a new application, we can call it a new technology.
Tsujimura: If you stop using technology, you should flag the reason. When the reason for stopping the technology is resolved and the technology can be used again, as long as the flag is up, it will not be forgotten.
Kashiwagi: There was a reason. It was too early.
Shingubara: It was fun to think back then. However, sputtered Cu has already disappeared.
Ohba: At that time, Cu patterning by dry etching was tried, simultaneously. They said there was no need Damascene process. It was tough to convince them from the reason of thermal budget.
Shingubara: I remember doing Cu-RIE at Toshiba for a while, I remember that.
Kobayashi: Each company was conducting Cu-RIE, and there were many papers on trace additives for Cu sputtering.
Shingubara: Did we do that in Japan?
Kobayashi: Plating was not done.
Ohba: Yes, I think so.
Moriya: When the technology node reached 1 µm, stress migration occurred rapidly, and Cu addition was discussed. Some time later, interconnect resistance became a problem. When that started appearing, Cu interconnect was introduced.
Shingubara: That is right. Therefore, it was in the latter half of the 80s. I think it was around 1988 or 1989, and Cu was quite popular in Japan.
Kobayashi: Did you perform Cu-CVD in the mid-1980s?
Ohba: We did it after the source was available, and tried many precursors such Cu (hfac)₂ and other sources. In addition, we tried Au-CVD at that time.
Kashiwagi: I have also performed CVD for Al. I think it was in 1970.
Shingubara: Were you performing this in the 1970s? That was very early.


7. Selective CVD: Picking up the failures

Shingubara: I think W is one of the very few successful examples, but how many years did it take to start the development in earnest, and then to use it in many places? I wonder how many years it took before it became really useful.
Kashiwagi: I cannot tell, after all.
Ohba: It took more than ten years.
Kashiwagi: Ten years. No, wait a minute, because W is....
Ohba: No, we look at history, it has taken over 25 years.
Shingubara: No, no, no, I am not talking about a paper, I am talking of when many companies started considering W in their technology development.
Kobayashi: I think W silicide was the earliest; however, that was about five or six years after the deposition machine was released.
Shingubara: That is five or six years for silicides? Therefore, it was quite a long time to embed contacts.
Kobayashi: Yes, long, long time. Was it finally blanket + etch-back? I was also working very hard on selective-W (laughs). That is what I was burning for...
Shingubara: Therefore, I would like to hear about the battle between Selective and Blanket. However, this is not a battle. I would like to know what you prefer.
Kobayashi: I am not sure about Selective-W. I still do not rely on it. But the selective deposition itself is now widespread.
Kashiwagi: It is widespread.
Kobayashi: For SD (source and drain) with SiGe. That is selective deposition. Moreover, as you probably know, strained-Si is a selective epitaxial material, coming in at the 45 nm level. Therefore, no solution was found for selectivity loss, etc. at that time. No breakthroughs were observed.
Kashiwagi: However, I think that the selective approach is, in principle, an extremely unstable technology. If there is a slight shift, it will explode. If the preprocessing is slightly different, it will stick together.
In addition, we have been investigating selective epitaxy since 1960, when we first started. There was chlorine in this type of thing. It is okay right next to the pattern but a little farther away from the pattern, it gets deposited. Finally, it is a matter of how much can be controlled by surface treatment and peripheral technologies. If it gets a little dirty, it sticks to the surface.
Kobayashi: It would be fine if it did not work at all after that, but as I said, Intel's technology was a real surprise. SiGe is selectively grown in the source-drain area. In the actual device.
Masu: That is because it is smaller.
Kobayashi: However, if a nucleus is formed elsewhere, even just a little...
Kashiwagi: If that happens, it is not good.
Kobayashi: If that were to happen, would the entire process be ruined?
Ohba: That is why the related technology such as gas supply system and clean process chamber has improved.
Tsujimura: I am not sure why or what, but would W work now?
Ohba: I think selective growth will be used widely work now.
Moriya: I think selectivity has probably increased. At that time, however, it was necessary to make it thicker by the selection, adding approximately 1 µm.
At first, we thought it would be good if we could make it at approximately half a micron; however, we were aiming for approximately 1 µm. If we did that, it would have been quite difficult to maintain selectivity.
Kobayashi: This is true. This was the case at the time. Therefore, that was reckless. Present applications are not a very thick deposition. Thus, it is possible to prevent large bumps in hemispherical grain (HSG) to a certain extent as far as the deposition thickness is limited.
Kashiwagi: I see.
Shingubarahara: Thus it is because it is smaller. I wonder if this is because it is smaller.
Kobayashi: The fact that thin deposition is required for certain applications makes a compromise. That is why I think technology is advancing, regardless of what we said in the past. Just because something fails in the past does not mean it is dead; however, if some people use the technology in different applications (materials), it will bring success. Selective epitaxial growth is a good example.
Kashiwagi: That is true when you go through an entire cycle.
Masu: The BOX separation was historically the first runner of STI technology. When Prof. Shibata (Tadashi Shibata-Toshiba) developed, the planer separation of BOX did not succeed due to the large stress of thick oxide. However, the thin oxide with small stress has succeeded as STI nowadays. We know that It becomes possible because it became smaller.
Kashiwagi: In my working environment back then, I now think it is a huge step forward.
Masu: Various vacuum systems such as preprocessing systems, multi-chambers, etc. have been introduced. The equipment vendors have greatly contributed to equipment stability.
Ohba: This is why every device manufacture asked for equipment vendors to develop individually. Meanwhile, I saw the Toshiba chamber and ours at the same transfer platform.
Kashiwagi: In the past, the -9 or -10 powers of 10 were the level of extreme research and ultra-high vacuum was meant for university research; however, it has become a normal production technology.
Ohba: There was a significant evolution in this area, such as the use of magnetic fluid seal system for ultra-high vacuum transfer developed commercially.
Shingubara: We could do what we could not do in the past when we try.
Kashiwagi: It might be possible. That is true.


8. CMP: Removable Flattening Revolution

Shingubara: Because Mr. Tsujimura is here, I would like to talk about CMP.
Tsujimura: I started in December 1989; hence, I think these two (Kashiwagi and Moriya) are more knowledgeable than I do. CMP has been used by IBM in reality since 1985.
Shingubara: Did you use it in 1985?
Moriya: That was in 1985 when a person from IBM came and told me that they were doing five-layer interconnects.
Tsujimura: I think IBM had a prototype line at that time and was doing various types of planarization.
Shingubara: Were you also working on metal?
Tsujimura: I do not know if they were actually used in devices; however, I think at least in terms of concept, oxide film, W, and Cu damascene were included from the beginning. We also had plating equipment for semiconductors from 1988, and IBM was interested in this equipment from that time. In the hindsight, I think that was the Cu Damascene. I was too busy thinking of CMP at that time.
In Japan, the adoption of the oxide film was relatively early; however, in 1994, etch-back was the mainstream, and I think CMP was strongly rejected. Cu was earnestly adopted in 1998; however, I think it was actually developed much earlier. On reaching this point, anything and everything is CMP for planarization. Novel applications, such as metal damascene gates and metal hard masks, are on the horizon. This is the sequence of events.


9. How to realize innovative technology

Ohba: I have a question. By overlooking semiconductor history, what is the root cause the innovative technology becomes the de facto standard? In the past, IBM drives it. However, there is a further reason not only technology but also economic situation. In the early 1980s during the economic downturn in the United States, a huge number of layoffs were conducted even at IBM but IBM recovered worldwide potential afterward. If an economic situation as Japan had a market share of more than 50% continues until the 1990s, do you think Japanese technology became the de facto standard?
Tsujimura: I think that IBM came up with CMP as an option. I think it will become de facto only when other device manufacturers, such as Toshiba and Micron, start using it. However, I think what is amazing about IBM is that they are the first to introduce not only CMP but also SOI, low-k, metal cap, and other revolutionary technologies.
Ohba: IBM only changes one challenge process. Before volume production starts, IBM tries to implement a challenging process before Two generations to prove with tangible results.
Tsujimura: I see.
Ohba: That is the smart strategy of IBM. Therefore, if it is SOI, they use SOI, for example, by using 0.18 µm node, which is unnecessary to implement. Those two generations are a kind of learning period to make sure processes. In Japan, they tried to implement many challengeable processes in one generation. As a result, most Japanese manufactures have to spend time and cost, which means it was low profitability.
Kashiwagi: However, IBM really did a great job in 1985 or so, I do not remember. How did they use CMP?
Tsujimura: I think they bought all the CMP equipment available at the time.
Kashiwagi: Previously, it was forbidden to touch the delicate surface. It was outrageous. Even for plating, a time existed when CVD was quite common. When I was at Applied, A guy from Bell Labs said that plating could not be used for LSI. Many people were like that.
Tsujimura: We have developed gold plating for clean room use since the late 80s. Notwithstanding, we did not think that Cu plating would be used for semiconductors. Is it not pathetic?
Kashiwagi: Gold plating was used for metallization to a large extent. However, it depends on the company; I still think that various problems occurred when Cu was introduced to LSIs from Al-Si to Al-Cu. From our point of view, the trouble was nothing but resistant. Al-Si-Cu has not been used in bipolar devices for a while, and I dislike it when LSI devices become MOS devices. This is not necessarily a kind of technical logic. They did not want to use Cu.
Shingubara: In the past, were there discussions on Cu contamination?
Kobayashi: We used to discuss a concern about it a lot; however, no one said concerning it anymore.
Kashiwagi: That is something we had a long preconception about. However, each LSI technology is an accumulation of experience, isn’t it? This is not just a matter of getting the physical model right. Rather than just trying it out, you have to go ahead before understand the entire process. That is why we have to be conservative, especially in the field. Must ask ourselves "Are you sure of this?” with the best and a few of our knowledge.
Shingubara: Polishing was also done slightly early on. Fujitsu started very early.
Ohba: They were involved in it.
Tsujimura: Did you also published a paper on Damascene, back then?
Shingubara: At Toshiba, I do not know if they published a paper or not, but there were lots of people who attested to it.
Kobayashi: Hitachi was also involved in it. However, I do not think they succeeded to apply it to interconnect or anything related to that.
Kashiwagi: Why did IBM do it and Japan did not? For example, Toshiba has been investigating STIs since the 1960s. It was just an idea. Why did not we consider our ideas and make them as persistent as IBM?
Ohba: It comes intellectual property of IBM according to business strategy. For instance, IBM prepares ten times solution to one problem and screens out candidates.
Kashiwagi: They have enough room…
Ohba: I think it was a business size. In Japan, the size of business is saturated and decreased due to the small Japanese market. As semiconductor needs huge investment at every generation, profit of Japanese manufacture became low and thus the activity of R&D became worse.
Kobayashi: I often hear such a story that a certain number of people develop similar technology at a similar time in global, even though I am not sure why, to a certain level.
Shingubara: It is hard to contine...
Kashiwagi: IBM, for example, has a good yield rate, or there are some areas where they hit the mark. I wonder what this boundary is.
Ohba: IBM always made a strategic plan in the global market every ten years. In other words, IBM made market direction intentionally according to their strategy.
Tsujimura: Is it possible for IBM to continue developing innovative technologies in the future?
Kashiwagi: It is hard to think that IBM will continue to produce any new manufacturing technologies in the future.
Kobayashi: I do not think either. As I earlier said, you have to warm up the old and make it ready for new mass production, and if you do not have a mass production line, you cannot raise it up to that final level. Mass production will be the final check.
Kashiwagi: If IBM fails to produce such a novel technology, where in the world do you think it will come from?
Ohba: I think there are very few novel technologies.
Shingubara: I do not think there will be any.
Masu: I hope that the era of three-dimensional devices will be coming.
Ohba: 3DI will come out, definitely.
(August 4, 2007)

Composition: Eiichi Kondo; Everyone spoke passionately. I enjoyed listening to their passionate approach to the development of multilayer interconnect technology. It means technology matures through repetition. If you feel it was just a while ago, it is a sign that you have lived in this field for a long time, including me. All names and affiliations in the text were as they appeared at the time of writing, and titles were omitted. As a rule, the affiliations of the interlocutors are as they were at the time of the interview, and " former" has been added.

Translation: Shinji Yokogawa; I am not sure if I have accurately translated your discussions at that time. But, I hope that the challenging atmosphere of those days, from birth to the maturity of the technologies, will be conveyed in full. I hope that young researchers and engineers will feel the joy of research and challenge.



   Previous Page    Go To Top

Notation based on the Specified Commercial Transaction Act