iSixSigma

Placket-Burman Designs

Six Sigma – iSixSigma Forums Old Forums General Placket-Burman Designs

Viewing 42 posts - 1 through 42 (of 42 total)
  • Author
    Posts
  • #40568

    Ivan
    Participant

    I have been performing DoEs using PB designs for initial factor screening purposes. Typically I will start with about 10 factors. I have been selecting 20 runs, even though 12 runs will do as per basic design.
    The reason I have choosen 20 runs is because I perceive it to be better than 12 runs in “diluting” the degree of saturation. Please advice me whether I am gaining significantly from this approach.
    My last step in DoEs, is full factorial designs – factors selected from screening DoE.
    Thanks.

    0
    #126216

    Mikel
    Member

    Since you are only looking for main effects and are willing to take the risk or have enough process knowledge to know the risk, you gain nothing except added costs by doing more than 12.
    You finish with full? – what about response surface designs?

    0
    #126217

    Ivan
    Participant

    Thanks. Stan.
    The cost implications is definitely an issue. But, by doing 20 runs I am gaining in terms of Quality of results to avoid greater losses at later stages. I employ alpha=20% to insure from risks.
    Without surface response, we have generally achieved process robustness (6-7 sigma). There has been no incentive to attempt furthur.
    Thanks.

    0
    #126218

    Deep
    Participant

    Ivan;
    You only need 11 runs to analyze the 10 factors per the basic design, not 12 as you mentioned. If you have money to spend on 20 runs why dont you go for fractional factorial experiment.
    Deep

    0
    #126219

    Ivan
    Participant

    Thanks. Deep.
    I would prefer fractional factorial. But IV resolution is minimum 32 runs. Its expensive compared to 20 runs.
    Thanks.

    0
    #126221

    Deep
    Participant

    Ivan;
    I think it is better to do a res III factorial design. The advangaes are, Aliasing structure is not messy like PB. Another thing is i dont think all the 10 factors that you have will be significant. If some of the factors are not significant you can remove those from the model and the design will get better resolution. Another good thing is if you want to filter out some main effects from the confounded two way interactions by foldover method. I think it is better to use factorial designs esp when you can afford to run 20 runs. Just my opinion. Please wait for other experts. There are real experts in this forum, Robert butler, BTDT, DaveS…
    Thanks
    Deep

    0
    #126222

    Mikel
    Member

    Deep,
    I don’t know of any 11 run designs – tell me about them please.

    0
    #126223

    Deep
    Participant

    Stan:
    I am searching for the book that has mentioned about PB design. I will give the details after lunch. I can tell this now, one good thing about PB designs are we can have n-1 factors, where n = number of runs.
    Thanks
    Deep
     

    0
    #126224

    Mikel
    Member

    You are kidding yourself about gaining in terms of Quality. An alpha of 20% also is not good insurance.
    But if you are really getting 6 – 7 sigma you ought to trademark and copyright what you do and forgoet about asking questions of idiots like me.
    BTW – I don’t believe you generally get 6 -7 sigma. This says you are running all processes defect free after your rudimentary DOE pass. Doesn’t match well with my experience.

    0
    #126229

    Deep
    Participant

    Stan and Ivan:
    Sorry , I made a mistake saying that in PB you need only 11 runs to analyze 10 factors. As Stan mentioned PB does not have a design with 11 runs. Only multiple of  4. Using the 12 runs we can analyze upto 11 factors, but it will be a saturated design. Sorry once again
    Deep
     

    0
    #126243

    “Ken”
    Participant

    Stan,Let’s twist a bit… In exploratory studies it’s not confidence or alpha that’s important. Instead, the Power of the Test drives our ability to see differences between trials. Given a fixed variation, and trial size–if I wanted to increase the power of my expermental evaluation what would I need to do with the alpha value for the test? I’ve done many studies, both passive and active, using alphas of 20%. I such studies, I wanted to insure I did not miss a key variable…Placket-Burman Designs are screening experiments for exploratory studies. Resolution IV designs are Characterization Designs that allow one to identify interactions separate from main effects. In these designs I recommend the use of a 10% alpha. In Full Factorials and Response Surface designs one should consider using alpha of 5% maximum, with 1% typical.Ken

    0
    #126246

    Mikel
    Member

    The alpha risk says I may choose to include a factor in further experimentation that I should not have. Given that I will likely reduce the model in the further experiment, the only real problem is that I have to control a factor in the further experiment that I really did not need to.
    So yea, I agree that it is okay.
    My point to Ivan is he thought 20 runs vs. 12 gave him insurance. Sample size gives him insurance, not # of runs.
    Hey could I have that Distinct Category spreadsheet you were passing around last week? Send it to @aol.com">StanMikel@aol.com or the other email that you have for me – it hasn’t changed.

    0
    #126248

    “Ken”
    Participant

    Stan,
    To be honest Ivan is on the right track.  Increasing the number of trials per design increases the power to detect differences between trials.  Within trial sample size has a much less effect on between trial power. 
    If you’ve ever looked at the DOE software package EChip this feature is emphasized during the DOE design stage.  You can experiment with trial size in Minitab, but unfortunately it’s not fun.  There is much in the literature on the effect of trail number and ability to detect between trial differences.
    I’ve passed out a number of copies of my spreadsheets to others with the only request that they provide me some feedback on its use.  Darth has a copy of the Distinct Categories spreadsheet, and I suspect you can acquire from it him.  If you are unable, then let me know.  Do you still have your place in Northern Illinois?
    Ken

    0
    #126250

    Mikel
    Member

    I know eChip, but you are wrong.
    You are talking about the difference between repeats and replicates, you don’t have to have more distinct runs to do that.
    If you want to send me the spreadsheet., I’d love to look at it, but I’ll not get it from Darth.
    Yes, I’m still north of Chicago and I remember where you live and that you offered to share some wine last time I saw you. Is the offer still good?

    0
    #126251

    “Ken”
    Participant

    Stan,
    I know the difference between repeats and replicates.  If I’m wrong, then please tell me specifically why I’m wrong.  Glad to receive some wisdom from the top guru!  That’s what makes this forum work…
    You know I’m interested in getting a new copy of eChip for myself.  It’s a great little piece of ware!  Do you happen to remember the name of the developer?  I believe his name is on the lead screen when you open up the software…
    Ken
    P.S.  Glad to share a bottle of wine together.  What’s your preference, white or red? 

    0
    #126257

    Buridano
    Participant

    Dear all ,
    I am not an expert like you are but PB designs can only accomodate a number of runs which is a multiple of 4.
    For ten factors it can only be 12 , the design must be generated as if one had to study eleven factors but one coloumn is not used.
    A fractional factorial down to resolution III can accomodate 10 factors in 16 runs and still keep main effects unconfounded.
    The usage of 20 runs is quite odd ( actually I don’t understand it ) as it adds extra runs/cost and still keep the same resolution even if the correleations between main effects and two-way interactions can be reduced. ( Not down to zero so confounding is still there )
    As far as I know there is no “quality” gain in the sense that the power of the analysis doesn’t change drastically.
    Another point I want to make is that screening designs ( PB ) are meant to screen the factors so I wouldn’t bother that much about the power once I can independently estimate my factors.
    Rgds 

    0
    #126265

    Darth
    Participant

    P.S.  Glad to share a bottle of wine together.  What’s your preference, white or red? 
    Knowing Stan, I believe you will have to substitute the word “or” with the word “and” along with some Jack & Coke, Tequila topped off with a little Grappa..

    0
    #126271

    Mikel
    Member

    I believe it is Wheeler (no relationship to Don).
    Red.
     

    0
    #126272

    Mikel
    Member

    Fractional Factorial can accomodate the number of runs -1 factors and keep main effects unconfounded. That would be 15 factors in a 16 run design. Try it in Minitab and look at the alias table.

    0
    #126274

    Mikel
    Member

    Good suggestion on the Jack. Start with shots and then go to the parking lot to find out why the condesending attitude since arriving back from Europe. I am assuming he spent a lot of time in France.

    0
    #126288

    “Ken”
    Participant

    Actually Stan, it was Austria. Assumptions can get folks in a lot of trouble. The French are not as bad as claimed. They have the same proud sense of nationalism as Americans. Otherwise, they look to be treated with a bit of dignity, just as we do! What you interpret as a condesending attitude by me is a reflection of your responses. I’m just attempting to mirror your behavior back. If you want to engage in a discussion, then let’s do so. But, an answer of “wrong” to a fairly complex topic is not something I’d expect from a guy like you. So, instead of floating snipes back via others on the forum, why don’t we discuss these topics openly. I tried that with Andy U until he got a little unruly. I know, as a business man, you won’t do the same to me…So, are you willing to join me in substantive discussion on this topic Stan? As I recall: I suggested that Ivan was on the right track by increasing the number of trials in the design instead of increasing the number of samples per trial per your suggestion. I suggested that doing this would allow Ivan greater power to see differences between trials. You suggested this advice by me was wrong. I said great! What’s wrong about it? Can you provide and explanation instead of a judgement? I would like to help, if possible. I’ll use one hint word–“Resolution.” Are you willing to pick up the ball from here, and run with it? Not trying to be condesending. Just trying to get this discussion forum to move from judgement back to discussion. I would like this forum return to being informative for those requesting honest advice of some of the more experienced. Are you willing to join me in this effort? I know you have the capability. Perhaps if you can show me some leadership I would be willing to sit with you across the table, and share a glass of wine. I no longer shoot shots of whiskey…Ken

    0
    #126289

    “Ken”
    Participant

    Stan,Ah yes, now I remember. Bob Wheeler if I remember correctly. Thanks for the memory jog.Ken

    0
    #126297

    Swaggerty
    Participant

    Ken,
    I’ve been reading your posts for the past few weeks and went back and read others from 2001. What I take from this is
    You are a knowledgable quality guy
    You have an interest in Six Sigma, but have never really implemented
    You went to Europe to do something other than Six Sigma
    You are now back as a Six Sigma consultant
    You are going to sort out the misfits like Andy U. and Stan.
    Interesting.

    0
    #126304

    George,
    Well done .. very astute.
    It’s clear that Ken has little pratical experience ..  as far as I’m concerned anyone who peddles Response Surface Methods falls into that category. Why? Because Taylor’s expansion assumes small deviations …  and that is not a good method for exploring process space! Although I admit it might be useful for investigating a mathematical emulation, such as Air Academy’s Airport, but its not much good for real processes.
    One way for Ken to demonstrate his superior knowledge is not by ‘book’ learning, but by posting a real case study. For example, how he personally solved a semiconductor problem … as many of us have over twenty-five years experience in that field, and I’m sure we would be able to ask some pertiment questions!
    We should remember, mathematical procedures are only models and real improvements come from ‘practical knowledge’ – knowing which methods actually work! Great detailed knowledge in one small area may baffle brains – it rarely fools nature!
    If I have a process surface in the form of an  ‘s curve,’ and have to ask myself where to operate. Clearly, I should avoid the steep part of the curve in the middle so there really is no point exploring minutae around the cusp,  or even around the ‘knee.’ I have to decide whether to put  my set-point … on the bottom flat part of the curve ….  or the top flat curve. Since in practice these parts of process space can be much further away from a ‘typical’ operating point – a mere +/- 10% variation in factor levels is not going to reveal much. This is why many companies prefer to use EVOP .. for example, one famous chocolate company uses the Simplex method, in a kind of controlled ‘tweaking.’
    Of course, none of the preceeding discussion takes into account the inefficiency of protecting again spurious interactions, the question of scaling, and many of the other issues now familiar to long term members of this forum. This is why I prefer to use OA’s as mentioned in the discussion with Michael Schleuter.
    As far as I’m concerned anyone listening to Ken is going to lose their shirt. I saw it in 1984 at Motorola and I’m seeing a lot more of it recently. Some operations managers I’ve met would give their eye teeth for a rolled yield of 95%, let alone a six sigma process! No wonder everyone is off to China!
    Andy

    0
    #126309

    “Ken”
    Participant

    George,
    Interesting assessment based on your observations of scant postings on the Internet.  We both know how reliable info on the Internet is these days!
    Again, contrary to your claim I’ve been in the change game for over 20 years.  The most recent chapter in this game is Six Sigma.  Contrary to your claim I led a larger implementation in Europe.  Unfortunately, the company I supported made a lot of bad decisions, and place my work, and the work of others, onto the back burner.  Perhaps you know this feeling.  It’s the same one you had with Motorola years ago…
    I do work as a Six Sigma consultant as I sort out the next chapter in my life.  I became a consultant due to a demand by colleagues and friends who sought my guidance and support as a change agent, and problem solver.
     I consider neither Andy nor Stan as misfits, contrary to your claim!  It would be nice if they decided to join in collaborative discussions, rather than push their particular slant and agendas on the greater forum community. 
    OK, George hopfully that answers both yours, and Stan’s questions.  My question is are you willing to join me in a substantive forum discussions, or will you continue throwing up these barriers to confuse clear communication?
    George/Stan, are you truly a leader on this forum–or, are you simply a player?
    Ken

    0
    #126310

    “Ken”
    Participant

    Andy,
    It’s too bad you have to say such things without any understanding of the facts.  You know as well as I that it’s difficult, if not impossible, to provide substantive examples of work on-line.  But, I’ll tell you that you will be able to see some of my work in the upcoming Pharmaceutical Engineering journal in September.  You see while you were dreaming of a chance to publish, I was doing it!  Much of my work will be on the cover of the next issue of PE.
    I’ve not seen any of your work, or examples in part displayed anywhere on this forum.  So, I’m not sure why you think I need to do the same to support my position.  I’ve interacted with at least 5 forum members off-line to provide pro-bono guidance in their projects.  All of them have told me my guidance is the best they’ve received yet. 
    OK, I’m sure that means nothing to you.  So, I’ll tell you what Andy.  Provide me with your email address via this forum.  Off-line I will request from you specific contact information that contains your name and address in Europe, I believe you’re in GB, and have you agree to a confidentiality statement.  If you do this thing, I will gladly send examples of my work that should quickly disprove your claims.
    Andy, you seem like an intelligent guy.  But, for the life of me I don’t understand why you persist in saying such unfounded things about someone you don’t know.
    You seem to think that I don’t understand some of the approaches you advocate.  However, I not only understand I’ve used many in the past, and some still.  You talk loosely about EVOP…  I’ll tell you what.  Follow up with my suggestion above, and I’ll show you some novel ways of process improvement that you can only dream of!
    Ken  

    0
    #126311

    Ed
    Participant

    Unfortunately, the company I supported made a lot of bad decisions, and place my work, and the work of others, onto the back burner. 
    You worked for a multi-national company and were fired?   Correct?    Says something about how effective you were?   Spout-off more.

    0
    #126313

    “Ken”
    Participant

    Ed,
    Believe what you would like.  We are all entitled to our opinion.
    Ken

    0
    #126314

    Ed
    Participant

    I do not necessarily believe anything that could be determined by my previous statement.   I was merely trying to start an argument.   There were no new releases at the video store and nothing on cable. 
     I heard that most of your statistical advice is narrowly derived and applied because your mother dated heterogeneous populations.

    0
    #126315

    Bob P
    Participant

    Ken,
    I’ve followed this thread closely, and have noted and interesting pattern.  Every time you confront Stan with a follow up explanation he cleaverly evades your request, no matter how courteous you make it.  I’ve noted the same response by your friend Andy U.  This leads me to wonder if both Andy and Stan are working together so how, if you know what I mean.  I suspect that Stan is a real lonely guy who has nothing better to do than sit at the computer and invent new screen names to give the real guys like you a hard time.  It’s a very sad way of chasing good talent away!
    By the way, good call with George being Stan.  I thought the same.  But, have you considered that Ed is also Stan.  You know, my thinking is that Stan or what ever his real name, is probably a real insecure guy that has problems with your kind of talent and skills.  Anyway, please hold your ground.  We need guys on this site who give real answers. 
    Thanks again with your help on properly scoping my recent project. 
    BP 

    0
    #126316

    Ed
    Participant

    Sure, Ken/BobPP, anything to draw attention away from your sister.  It must have broken your mother’s heart when sis started her stochastic modeling wearing only those big tacky gold Markov chains, taking those long high heeled random walks only to wind up grinding away in her stationary process, thugs lined up around the block in an oblique impaired resemblance of queuing theory, and for her only to have wound up pondering birth and death processes with their inevitable Poisson relief.   Hopefully dramatic intervention is just around the corner for you Ken/BobPP, Mom and Sis.   Our prayers are with you. 

    0
    #126317

    “Ken”
    Participant

    Stan,
    I hear you.  It’s a rather boring night without something on the boob tube.  I think I’ll take a brisk run to work out the kinks instead.
    You know what they say in my part of town–“Narrowly derived statistical advice is better than no statistical advice at all…”
    It’s been fun!  Please be sure to tuck Andy into bed, and tell him a good bedtime story for me.  We both want to insure that he is well rested for his responses tomorrow!  I’m sure the entire forum will be ready for our show tomorrow.  When I get back from my run, I’ll get to work on my entertaining responses to you, Andy, and all of his friends.  The forum has got to get it’s money’s worth, right?
    Ciao,
    Ken
    P.S.  Stan, I think you are a master of diversion!  You’ve managed to push our original discussion completely out of the field.  Bravo!   We should invent a new role in Six Sigma, MBB-Master Bilk Belt.  I’m certain you qualify.   I wonder if you’re this masterful with your customers…  As always, it has been a hoot!

    0
    #126318

    “Ken”
    Participant

    BP,
    Thanks for the follow-up.  Appreciate your response, and kind feedback. 
    Yeah, I was aware of what was going on…  Stan is both a busy and funny guy.  You’ve got to admit he’s got a great sense of humor!  However, I’m really wondering if Stan is still in business if he’s got all of this time to play around. 
    Glad I could be of help with your project.
    Ciao,
    Ken

    0
    #126319

    “Ken”
    Participant

    Eddie/Stan/Andy/George,
    You know, maybe we need to get together after all.  Imaginations like ours should be shared over some wine, whiskey, or other spirits-and possibly some good talk! 
    I’m good next week on either Monday or Tuesday for lunch.  What do you say we get together at the old Road House on 83.  Let me know the time, and I’ll be there.  Feel free to bring Amy if you like, but I think we will have more fun mano a mano.  Let me know.
    Ken

    0
    #126322

    Ed
    Participant

    We’ll get back with you on it.  You’ll have to be sure to bring your stat books and pseudo-academic references so you’ll have something to talk about.   I’ll look for the dweeb with the textbooks and post-notes. 

    0
    #126324

    “Ken”
    Participant

    Good enough Stan! Again, enjoyed the fun and games. Ciao,Ken

    0
    #126333

    Ken,
    I already gave reasons for my suspicions. I also gave reasons why RSM is flawed. You’ve chosen to ignore both. I’m not even asking for a blow by blow account of one of your projects – just describe one, in layman’s terms, for the benefit of the forum. Unfortunately, I can’t comment on a pharmeceutical application because I’ve no experience of this industry, but I would love to read a semiconductor war-story. I’ll even follow yours with one of my own!
    In summary, I believe RSM is a worthless procedure and anyone recommending its use has little experience of real processing. (I make no other claims about your other areas of expertise, but I admire your willingness to help others …)
    I also agree that sometimes Stan is too curt .. see I have no problem criticising something he says either, and he would be the first to jump on my case if he disagreed with me. That’s why I respect him and his considerable practical knowledge. Yet, I have never met Stan, or anyone else in this forum. Being rude is all part of a robust debate, in the best Graeco-Roman tradition. Now that you’ve overcome your over-exaggerated sense of politness, your posts make a much more interesting read.
    Andy
     
     
     
     
     
     

    0
    #126334

    “Ken”
    Participant

    Andy,
    RSM is a worthless procedure…  Strong statement…  I’ve used RSM successfully in many of my past projects within a manufacturing setting.  I believe I remember you suggesting something about variation transmission being a better approach…  Perhaps I’m confusing you with another intelligent engineer.  Anyway, yes I agree with your statement Robust Tolerancing, aka. Variation Transmission Anslysis, is more efficient than a Dual Response Approach using RSM.  However, before you can do VTA you need a fairly well developed means model of the process.  Something only RSM can deliver.  Can you do VTA without a good model?  Yes, but you know the saying–garbage in, garbage out!
    So, no pharma or med device example.  Only a semiconductor example, else you won’t understand.  I guess there is no way to draw any learning from other industries on your side of the pond.  That makes my job quite challenging.  I’ll need a little time to search through my records.  It’s been a while since I worked in the semiconductor industry.  To give you a sense, when I was growing silicon crystals at Wacker Chemitronics I worked with 2″(50mm) diameters, and later graduated through to diameters of 4″(100mm).  Geez, that sounds like a long time to me.  I suppose you’re now working with 10-12″ diameters, but not sure.  I would be interested to know where the industry on your side of the pond is at now.
    To give you another reference, we had just broken the 1 micron gate length in the processing area when I was in the middle of my semiconductor tour.  Here’s a real wierd perspective to chew on, much of my time in the semiconductor buz was spent working with very strange materials like Indium-Phosphide, GaAs, and Germanium.  Yes, Germanium.  A real wierd reversal in technology.  You see Andy, I worked for government defense agencies on my side of the pond while in the semiconductor biz.  Germanium, as it turns out, when operated at close to cryo temperatures produces electron speeds 3X faster than silicon through the gates at the same temp.  Also of note is that Germanium circuits are much more radiation hardened than similarly configured silicon circuits.  Of course, that is if you used antiquated Integrated Injection Logic, an early Bipolar technology.  I was working to advance Germanium technology to MOS design and processing that would allow radiation hardened circuits to be developed.  Yes, you say that old news.  Yes, I agree today it is.  But, back in the late 80’s it was cutting edge stuff.  Here’s an interesting fact you may not know–Germanium does not produce a native oxide like Silicon does.  Instead, I worked hard to develop a process of Germanium nitridation.  Germanium nitride is native and produces the lowest interstitial states allowing MOS designs in Germanium.  Doing this allowed me to characterize CV plots similar to that of silicon.  This is all crazy stuff, huh?  You working with Ion Implantation at your location?  Here’s a blast from the past–I worked on a team to develop a direct write implantation system back in the early 90’s.  This was before it was written up in the journals for commercial use.  Imagine that back in the 90’s we were writing circuits into Germanium without the use of a mask.  Were you using e-Beam write systems at your factory before you left?  Wild ah!  Boy, this stuff seems like yesterday to me.  It was all such challenging work, I could go on for hours talking about it.  At the time, I worked closely with a team of scientists that had credentials as long as your arm.  In the early 90’s one of my colleagues who graduated from Carnegie-Mellon was developing a biological circuit to interface brain tissue with digital circuits.  Wierd stuff huh!  Can you say Borg! 
    Perhaps you remember that crazy initiative back in late 80’s the US government was talking about called SDI, AKA Space Defense Initiative.  Well, I did a little work in this area.  No doubt you can see why we wanted radiation hardened circuits that could process signals 3 times faster than silicon.  Well I’ll stop the boring essay.  I suppose you are probably not impressed with any of this stuff.  Anyway, I’ll need to look over my limited materials to see if I can find anything that will be understandable on your end that is not still held classified in one form or another.  You know, during that time I think I may have done some early work using Shainin’s methods.  Can’t guarantee anything, because I may not be able to find it.  But, if I can I suspect you should be able to understand it…
    So Andy, don’t hold your breath on me getting back to you with that exact application that you will need to understand my work!  However, I’ll do my best after the holiday.  I’m guessing your background is Electrical Engineering.  Worked with many EE’s in the past.  All good folks, but quite limited in stretching their minds past their noses.  However, I’m sure your different.  You have to be.  After all, you’re still pushing Shainin and Taguchi methods.  I surely want you on my side on this forum.  After all, without your santioning, I’m nobody!  Remember, it’s all in fun..
    Cheers,
    Ken

    0
    #126336

    Ken,
    I  worked on FLIRs many moons ago, so I’m somewhat familiar with Ge. Actually, Nitride passivation was first developed at the University of Surrey – they used it to encapsulate GaAs, which tends to dissipate As during an anneal cycle! But of course you guys invented everything …
    Ion Implantation – do you remember Hans Glavishnich (spelling?) He discovered the ‘tiger pants’ effect on blanket implants. Direct write molecular beams – I bet that beats the c**p out of the cycle time, but the ‘resolution’ must be fantastic. I bet de Broglie was impressed.
    I know little about CV measurements …
    I’m not sure, but I think the reason why the Ge/N interface has less interface states than Si/O is because there is a better molecular size match. How do you measure the interface charge from the CV measurement? I assume it’s difference to the fixed oxide charge :-)
    No I’m not an EE, but I don’t want to distant myself from nose stretching …
    Cheers,
    Andy
     

    0
    #126341

    “Ken”
    Participant

    Andy,
    Did you do any work with GaAs?  As it turns out I’m aware of some folks doing VLSI work and even some ULSI work with GaAs around the 90’s, but I only heard of it…  It would be interesting to actually talk to someone who did the work.
    The focused Ion Implantation work was real challenging.  In fact, I supported the characterization efforts on the early development.  Having been out of semiconductor work since ’91 I wonder how far the technology has gone today, or even if its still alive.  I’ve lost track of its activity.  Have you worked with implantation technology in the past?
    Capacitance-Voltage measures are typical for characterizing MOS process performance.  CV plots are highly sensitive to infacial charge build-up, and therefore provide a key measure of performance and potential threshold voltage in MOSFETs.  The technology works the same way with Ge, except the voltage profile is a bit different. 
    You worked with FLIRs also!  Good deal.  The starting material for infrared detectors has to be very pure.  Typically, the purity is measured in the ppb range.  One of my first research projects was the development of starting material for both IR sensors and celestial solar cells supporting a high altitude infrared optical systems.  Both require very high purity silicon starting material.  The material from Wacker and Monsanto at the time came in at the 10-100 ppb impurity range.  The solar cells were configured a bit differently than terestrial cells, but let’s not get into the mud.  In developing these guys I created a variant crystal growing process different from the two primaries, Czochalski and Float Zone, called Cold Crucible.  In the CC process the silicon was contained within an RF magnetic field on all sides, and supported on the bottom with water cooled pedistal.  The process was originally developed by Dr. Joseph Wenkus of Ceres corporation to manufacture Cubic Zirconia.  I adapted Dr. Wenkus’ process to silicon with remarkable results.  It was an amazing site to see molten silicon siting within an RF coil without anything holding it up except an invisible magnetic field.  Gee, those were some fun times…
    Have have you done any work supporting Oxygen gettering of impurities in MOS stuctures?  That was another fun piece I spent time on while in your industry.
    OK, it’s been fun–but, I think it’s time for breakfast.  I’ll keep looking around for that special project to demonstrate my Six Sigma abilities in the semiconductor area.  Are you sure I can’t use a more recent project supporting the Pharma or BioPharm industries to give you a flavor of my work?  It would make my life easier a bit easier. 
    Cheers,
    Ken

    0
    #126344

    Robert Butler
    Participant

      Screening designs are set up for the purpose of checking for simple linear effects of variables you either know or suspect have a major linear impact on your process.  They are built so that the effects in question are clear of one another but not of two way and higher order interactions. 
    Choosing to “dilute” the degree of saturation by just randomly adding experiments to a design will waste effort and buy you very little in the way of process/variable understanding. 
      If your variables are continuous and if you desire more information than can be gained from the basic screen then the first place to add experiments would be the center.  The basic screen plus two center points buys the following:
    1. Enhance the estimate of ordinary process variation.
    2. Provide a check for curvilinear behavior over the range of the variables of interest.
      While the second will not tell you which variable(s) exhibit such behavior the fact of its existence will indicate the need for more than two levels in your follow up design work.
      If you have a good reason for suspecting the existence of one or two interactions and you would like to check them with the initial design then you should take your existing screen and run it through an optimizing routine (most of the DOE packages have the ability to do design augmentation using D optimal techniques) in order to identify the specific experiments needed to separate your main effects from the interaction(s) of interest.
      The final design will be “optimum” from the standpoint of providing a check of main effects, presence of curvilinear behavior, check of one or two suspected intereactions (all clear of one another), a reasonable estimate of ordinary process variation, and for the situation you described will probably result in a final count of experiments which will be less than 20..

    0
    #126349

    Ken,
    No the guys working on ‘nitride encapsulation of GasAs (silicon nitride) used my equipment for measuring ‘hopping conduction.’ I believe there are still some groups working on direct bandgap materials in the UK.
    I studied II and CV measurements at Siemens Ballanstrasse, Munich. My understanding is that II is still used for Vt adustment and to form self-aligned gates.
    There are at least two other posters with a pharmeceutical background. The closest process to these are likely electro-chemical deposition and ink development. So you should not hold back on my account ….
    As you know, one of the problems with ‘raw materials’ is the question of ‘activity.’ This means processes such as ink development have to be adjusted, to compenstate for differences in chemical activity. Now it is often the case that the response is multivariate, which means that adjustments have to be made in Gram Schmidt space.
    It was while studying this problem that I gained a greater understanding of TM, which I’ve since shared with the forum.
    Cheers,
    Andy

    0
Viewing 42 posts - 1 through 42 (of 42 total)

The forum ‘General’ is closed to new topics and replies.