When I met Dieter Rathei several years ago at an Advanced Semiconductor Manufacturing Conference, we hit it off pretty quickly.
At the time, he worked on the yield team in one of the European fabs. Dieter's a very nice guy, boyishly handsome, whip-smart, and a bit irreverent, with an endearing mischievous twinkle in his eyes. But I sensed a certain restlessness in him during our conversations, an air of "been there, done that, could do it better" when he still toiled as a fab guy.
Dieter took the entrepreneurial jump in March 2005, starting a company in his native Austria, DR Yield. No, it's not "Doctor Yield," just a fortuitous branding coincidence with his own initials. The company operated in stealth mode for awhile, so Dieter couldn't share anything with me "on the record." But within the past several months, DR Yield has emerged from its self-imposed quiet period. After I received the recent press release touting the yield control software firm's exclusive distribution arrangement with Marubeni Solutions in Japan, I contacted Managing Director Dieter to see if he would agree to an email interview. Here are the Q's and A's:
Why did you start up this company?
I worked for 10 years in yield management in several different fabs, with products spanning from DRAM to advanced logic to analog/mixed signal. I found there is a need to put yield better under control, which can be achieved by thorough monitoring of the electrical product test data. There is a lot of information in these data, but this information is hidden in the large data volume. So a specialized software solution was required, but not available on the market. That’s where the idea to start up the company came from.
Where did you get your funding?
The company was funded by the founders, with a “business angel†investor involved.
What have you discovered about life as an entrepreneur, about running your own company, that you didn’t know when you started?
Since we started the business in the field of yield management, in which we have thorough experience, there were not too many surprises. But being interviewed, for example, is something new to me.
Why has the issue of parametric/systematic yield losses become so much more important?
I think the industry has put a lot of effort into the control of defects and random yield issues. On the other hand, the problems with parametric yield issues have become more prominent, because they are basically problems of variability. Even if tighter process control allows offsetting some of the variability issues, process variability and hence parametric issues are increasingly yield-limiting.
Given the influx of many new materials and processes into advanced chipmaking, with the resultant integration and yield issues, are there any chipmaking applications where the issue of random defects has become worse, or at least has not been overshadowed by parametric defects?
As particle sizes have a certain distribution, it is likely that the next generation of technology nodes will have to face “ultrafine particles†that may contribute to another increase in particle-related defect densities. Ultrafine particles are basically large clusters of atoms that have a mode diameter of 18 nm. But still, any problem arising from them will only be on top of the parametric yield issues.
What are the key manufacturing benefits of your initial product offering, YieldWatchDog?
YieldWatchDog monitors all electrical test data---especially wafer sort data---at the parameter level. It will give a warning if there is a suspicious pattern or trend at any parameter that may cause a yield issue. One of the core benefits of YieldWatchDog is its ability to handle the enormous amount of data that are generated at electrical test. A key advantage is the statistical data aggregation that we have incorporated which is tailor-made to the semiconductor manufacturing process. Thus, YieldWatchDog identifies the early warning signals of yield degradation before they become yield-limiting. It allows yield engineers to act on an emerging problem, instead of reacting to a yield issue.
What about the economic benefits. What kind of money can a fab or company save if it employs YieldWatchDog?
One single yield degradation that can be avoided or diminished will save many times the amount of the investment in YieldWatchDog. We published a paper together with a customer last year at the ISSM conference: the cost-savings potential of one specific avoided yield degradation was quantified there with almost a quarter-million dollars in one week. Of course, these numbers scale with the size of the manufacturing facility.
Does YieldWatchDog complement standard YMS approaches, and how?
Yes, if there is a yield management system in place, we use the YMS database as the primary data source for YieldWatchDog. Every few minutes YieldWatchDog is then fed with the most recent data out of the database. It is hence more of an add-on to an existing YMS system that creates additional value. However, if a customer doesn’t have a YMS system, we can also feed YieldWatchDog with data directly from the testers.
Does YieldWatchDog work with design-for-manufacturing approaches, and how?
We are looking into this. We develop additional modules based on customer requirements, and we recently got an inquiry about linking YieldWatchDog to a DFM solution. But I can’t tell you more about this here.
Does YieldWatchDog work better for certain types of chip-manufacturing applications than others (logic, flash, DRAM, ASIC, etc.), or is there more of a demand for it from certain types of chipmakers (foundries, IDMs, fab-lite)?
YieldWatchDog looks at all kinds of data: It can watch fail-count statistics on memory devices as well as analog test data on an ASIC. It is a valuable tool for foundries, IDMs, and fabless customers as well as for back-end test facilities.
Does it work well for both mature and cutting-edge fabs? How scalable or retrofittable is it?
The design of YieldWatchDog software and its configuration options ensure that it works well for both cutting-edge and older facilities. Cutting-edge facilities may want to use it as one part of their advanced process control system, while more mature fabs that do not want to invest in all flavors of APC can use it as a stand-alone process control solution. This especially makes sense for the manufacturers of analog/mixed-signal devices, which typically operate depreciated fabs and are always facing the challenges of parametric yield issues. In terms of software design, YieldWatchDog is completely scalable in order to cope even with the data flow from the largest facilities.
You mention fabs/foundries, TAP, and fabless companies as potential customers. What about equipment or materials suppliers, any benefits there?
The strength of YieldWatchDog is the ability to cope with the enormous amount of data that is delivered by the automated test systems. Sometimes we get inquiries from other industries, but we simply feel deploying YieldWatchDog technology there is like “shooting sparrows with cannons,†as a German proverb says. There would be the theoretical consideration of aligning with a manufacturer of test equipment, but we have not pursued such an option.
How many systems are in the field---what is your installed base?
We have one official reference installation at Renesas’s Landshut facility in Germany, and we are talking to several other manufacturers, where we have demonstrated the value of the system. But as a provider of an innovative solution, we are still looking for “early adopters.â€
From what you’ve seen thus far, are there areas where you see a need to improve YieldWatchDog, and what are they?
There is continuous interaction with customers: Sometimes there are small usability requests which we incorporate into the next release; sometimes out of this interaction ideas for whole new modules were born.
Do you have any competition; if so, who? If not, why not?
The hardest competition that I know of are efforts that some manufacturers have undergone in the past to create a similar control system. Some tried to tinker with their own control system for the test data, and because these homemade systems had poor performance records, they are now very skeptical whether such a system can work at all. But we provide the whole solution, and we have put many person-years of experience and development into it. We have patents pending for many of the key algorithms in our software.
How soon do you expect to receive approval of the patents that are pending?
This question must be addressed to the USPTO.
Do you have any other products in the pipeline, and if so, can you discuss them in general terms?
Besides of the continuous development of additional modules for YieldWatchDog, we are just starting a feasibility study for another revolutionary product for yield engineering, which we intend to launch two years from now.
What is your game plan for the next 18 months?
We are working on establishing YieldWatchDog installations in all main geographical areas of semiconductor manufacturing: The distribution agreement with Marubeni Solutions is a big step into the Japanese market, but we already have local distributors in the U.S. and Singapore and hope to add more local distributors soon, in China and Taiwan for example.
|
Blogs 
