SD1 Price is Sigma's smart PR move.: News & Rumors Talk Forum: Digital Photography Review (2024)

rf-designwrote:

I opt for "b2"

release with current bad yield and try to fix the issue and later slowly adapt the price to final target.

A yield issue should not necessary exist because of the three layered diffusion.

Not necessarily, but quite probably. The Foveon arrangement differs from the standard three well CMOS in two important ways. 3 well CMOS implants N wells in a P- substrate, and P- wells in an N- well in the P- substrate as you say. The Foveon chip instead implants an N-well in the P-substrate, a P-well in the N-well. Thus far it is the same as the standard process. Then it goes on to make a further N-implant a in the P-well. That is an extra ion implant, and makes the process different. (I know most CIS processes have extra implants, but my supposition is that the one Foveon needs is not the run of the mill) Further, the depth of these wells has to be precisely controlled, and as designed, since on the depth depends the colour filtration properties. I have no idea how accurately the implant depth is controlled in standard processes, simply because I've never had to know, but those designing Foveon sensors do have to know, and to ensure that it is accurate. Further, The Pixel design is complex. This is from their patent US5965875,

'COLOR SEPARATION IN AN ACTIVE PIXEL CELL IMAGING ARRAY USING A TRIPLE-WELL STRUCTURE' (Apologies for shouting, that's as copied off the patent)

(http://www.google.com/patents?id=CxcZAAAAEBAJ&zoom=4&dq=inassignee%3AFoveon&pg=PA1#v=onepage&q&f=false)

On my monitor the pixel is 13cm square, and the poly width is 0.5cm, or a ratio of 26:1. Assuming that the pixel is 4μm, that makes the poly width 150nm, and lambda 75nm, if my arithmetic is correct, not far off the limits of a 110nm process. Further, when I look at the well layout, I can't make it comply to the design rules I have access to (although they don't envisage this configuration) - so I suspect, unless the newer pixel designs are very different (which is possible) that this is very demanding in terms of accuracy of the ion implants, and if they go wrong the photodiodes are simply going to get shorted out. If the implant depth goes wrong the colour performance will be all over the place. I just think there are probably reasons why the old sensor stuck at 4.7MP, and why nearly quadrupling the pixel count (albeit on a slightly bigger sensor) was a big ask.

I do not know the required 3-layer diffusion profile but there are many standard processes below the 0.11u CMOS from Dongbu which provide a 3-layer diffusion (N+P-ISOWELL/NWELL on PSUB). It could simply be a parametric or functionan yield issue somewhere.

I don't think they can use a standard process, I think it needs a custom process to get the implants precisely as required, in placement and depth, and I think it makes big demands on that process.
--
Bob

bobn2wrote:

Further, The Pixel design is complex. This is from their patent US5965875,

The new sensor patent is US 20100155576. The top layer now has four sensors (called blue), with the middle and bottom layers having one sensor each (green and red).

http://www.freepatentsonline.com/y2010/0155576.html
http://www.freepatentsonline.com/20100155576.pdf

Bob

I'm just sitting here trying to absorb what little I can from you and Josephs technical posts. I have followed you and Joseph for quite a bit and at first could't comprehend but 10% of it ....but now am up to 50% comprehension!

I could never post on you guys threads due to my lack of technical knowledge but wanted to tell you both that I appreciate the effort you both put into your sometimes encyclopedia posts......even to type all that technical data and the defense of it would be a huge effort to me.

Anyway thank you both, I will keep reading and hopefully learn more about how those magical sensors work and are made.
Boris
--

http://public.fotki.com/borysd/

Or at least left me open to the possibility.

Where's Dr. Andrevez Rusek when I need him?

bobn2wrote:

rf-designwrote:

I opt for "b2"

release with current bad yield and try to fix the issue and later slowly adapt the price to final target.

A yield issue should not necessary exist because of the three layered diffusion.

Not necessarily, but quite probably. The Foveon arrangement differs from the standard three well CMOS in two important ways. 3 well CMOS implants N wells in a P- substrate, and P- wells in an N- well in the P- substrate as you say. The Foveon chip instead implants an N-well in the P-substrate, a P-well in the N-well. Thus far it is the same as the standard process. Then it goes on to make a further N-implant a in the P-well. That is an extra ion implant, and makes the process different. (I know most CIS processes have extra implants, but my supposition is that the one Foveon needs is not the run of the mill) Further, the depth of these wells has to be precisely controlled, and as designed, since on the depth depends the colour filtration properties. I have no idea how accurately the implant depth is controlled in standard processes, simply because I've never had to know, but those designing Foveon sensors do have to know, and to ensure that it is accurate. Further, The Pixel design is complex. This is from their patent US5965875,

'COLOR SEPARATION IN AN ACTIVE PIXEL CELL IMAGING ARRAY USING A TRIPLE-WELL STRUCTURE' (Apologies for shouting, that's as copied off the patent)

(http://www.google.com/patents?id=CxcZAAAAEBAJ&zoom=4&dq=inassignee%3AFoveon&pg=PA1#v=onepage&q&f=false)

On my monitor the pixel is 13cm square, and the poly width is 0.5cm, or a ratio of 26:1. Assuming that the pixel is 4μm, that makes the poly width 150nm, and lambda 75nm, if my arithmetic is correct, not far off the limits of a 110nm process. Further, when I look at the well layout, I can't make it comply to the design rules I have access to (although they don't envisage this configuration) - so I suspect, unless the newer pixel designs are very different (which is possible) that this is very demanding in terms of accuracy of the ion implants, and if they go wrong the photodiodes are simply going to get shorted out. If the implant depth goes wrong the colour performance will be all over the place. I just think there are probably reasons why the old sensor stuck at 4.7MP, and why nearly quadrupling the pixel count (albeit on a slightly bigger sensor) was a big ask.

I do not know the required 3-layer diffusion profile but there are many standard processes below the 0.11u CMOS from Dongbu which provide a 3-layer diffusion (N+P-ISOWELL/NWELL on PSUB). It could simply be a parametric or functionan yield issue somewhere.

I don't think they can use a standard process, I think it needs a custom process to get the implants precisely as required, in placement and depth, and I think it makes big demands on that process.

I've done thought experiments on some of those demands, like skew causing the "Italian flag" tint effect some users have reported.

I wonder how "bad" a sensor has to be, to be part of this "yield" problem. For all we know, each wafer is yielding 50 "grade C" sensors, and in the process of getting their first 100 decent camera sensors, they accumulated 1000 that they shipped off to Alt Vision for industrial customers.

If it's problems with skew or diffusion depths, I can picture making a "weird color map" for each camera, the way cameras have hot and dead pixel maps now, and utilizing sensors that test way out of spec in color uniformity.

Ironic. According to early Foveon literature, this was going to be the world's most cost effective sensor. I think the whole competition, Aptina, Canon, Kodak, Nikon, Panasonic, Samsung, Sony, could lay down a 30mp 16x24mm sensor that would match the 15mp 3 layer 16x24mm, and have it out the door in a camera, "concept-to-customer" as we used to say, in 9 months flat, and beat the original $1,700 Sigma target.

gaussian blurwrote:

bobn2wrote:

Further, The Pixel design is complex. This is from their patent US5965875,

The new sensor patent is US 20100155576.

Wow, thanks for that, I'm on it!

The top layer now has four sensors (called blue), with the middle and bottom layers having one sensor each (green and red).

http://www.freepatentsonline.com/y2010/0155576.html
http://www.freepatentsonline.com/20100155576.pdf

Now having looked at it, I'm pretty sure that this isn't the SD-1 sensor, This isn't an X3 sensor, as you say it has four (big) blue photodiodes and a little red and green one. The anodes of all the photodiodes connect to ground, which isn't the case in an X3 sensor 9and can't be the case, due to the layering of them. It's very possible that the patented invention is used in the SD-1 sensor. The major claim is:

1. An array of CMOS pixel sensors, each pixel sensor associated with a row and a column of the array and having a plurality of photosensors per pixel location, each photosensor coupled to a sense node through a select transistor having a select input, each pixel sensor including a reset transistor coupled to the sense node and having a reset input, an amplifier coupled to the sense node and a row-select transistor coupled to the amplifier, wherein:

the select inputs for pixel sensors in a pair of adjacent rows are coupled to select signal lines associated with the pair of rows;

the reset inputs for pixel sensors in the pair of adj acent rows are coupled to reset signal lines associated with the pair of rows;

the amplifier transistors in individual columns of pixel sensors in odd rows are coupled to a first column line through an odd row-select transistor having an odd rowselect input and the amplifiers in the individual columns of pixel sensors in even rows are coupled to a second column line through an even row-select transistor having an even row-select input; and

the odd and even row-select inputs for pixel sensors in the pair of adjacent rows are coupled to a row-select line associated with the pair of rows.

Which has nothing to do with the pixel diagram shown, which is just an example embodiment. I guess they've chosen a very weird and wonderful sensor to illustrate the patent to throw people off the track.
--
Bob

Joseph S Wisniewskiwrote:

Or at least left me open to the possibility.

Wow, that's one for the diary

I don't think they can use a standard process, I think it needs a custom process to get the implants precisely as required, in placement and depth, and I think it makes big demands on that process.

I've done thought experiments on some of those demands, like skew causing the "Italian flag" tint effect some users have reported.

I didn't know that those had been reported, but if they had, now we can explain them.

I wonder how "bad" a sensor has to be, to be part of this "yield" problem. For all we know, each wafer is yielding 50 "grade C" sensors, and in the process of getting their first 100 decent camera sensors, they accumulated 1000 that they shipped off to Alt Vision for industrial customers.

It's part of the bind they're in. Of course a sensor has to be a lot better for a $9000 camera than it would for a $1500 camera, which in turn is better than one for a $500 camera. That could have been the other strategy, put the grade C's in a $500 SD-1000 but it doesn't fit well beside a $9000 camera.

If it's problems with skew or diffusion depths, I can picture making a "weird color map" for each camera, the way cameras have hot and dead pixel maps now, and utilizing sensors that test way out of spec in color uniformity.

That is another possibility, they are getting a yield of the sensors, but each camera has to be individually colour mapped to get it to 'standard'. I can imagine that could end up costing a lot and restricting supply. Of course, the accountants will tell them to price just at the level where demand equals supply, and if each camera is individually fettled by someone expert in 'weird colour mapping' the supply could be something like one unit a day. If they figure out an automatic way of doing it, then supply will go up and the price might go down.

Ironic. According to early Foveon literature, this was going to be the world's most cost effective sensor. I think the whole competition, Aptina, Canon, Kodak, Nikon, Panasonic, Samsung, Sony, could lay down a 30mp 16x24mm sensor that would match the 15mp 3 layer 16x24mm, and have it out the door in a camera, "concept-to-customer" as we used to say, in 9 months flat, and beat the original $1,700 Sigma target.

See Also

OSRS Giants' Foundry Guide (Easy Follow-Along Guide)Bigby Presents: Glory Of The Giants – Giant Foundling, ExplainedPeering at the Past Ola and Per Appear in Print, Film and BronzeNvidia Vs. AMD Vs. Intel: Which AI Stock Is Best As Competition Heats Up?

Without doubt. They have the pixel designs (3.6 micron by my calculation), they have the readoff designs and once you have that, I wouldn't think even 9 months. Sounds like a day's work with a layout editor, and one test run just to check it works as expected (and there is no reason why it shouldn't.). The biggest question would be fitting it in the fab schedule.

bobn2wrote:

TomFlawrote:

bobn2wrote:

TomFlawrote:

SNIP

If Sigma really did proceed with a business plan so weak that chip price increases were as large as bobn2 claims, at least to me, that says a lot more about the business side of sigma than it does about the technical side.

I agree, to a point, but they were trying a highly innovative product, and innovation carries risks. My speculation as to what happened is almost all about the business side, not the technical side. One can imagine the sequence of events

SNIP

In any case to proceed knowing there was not a reliable chip source at a reasonable price defies any logic, no MBA would allow this to happen.

MBA or not, you are prognosticating about a business you know nothing of. Foundries sell a service. 'Reliable' means they deliver the semiconductor process accurately to the agreed specification. If the customer's design fails to work using that process, it is the customer's liability. The low risk route is to design conservatively, according to the foundries design rules, but as I explained, there may well be reasons that the Foveon engineers would not have done this. The question is, when would the Foveon management come to understand that the yield was below expectations.

These long winded speculations about what may have happened ignore the reality that no business would proceed with a business plan under the postulated circ*mstances.

Many businesses have - yield problems are reasonably common. Even Intel, the world's largest semiconductor company has suffered them, and failed to produce chips to its expected specifications. Read up on the whole NetBurst fiasco. A 'business plan' does not produce working chips, relentless debugging and yield characterisation does, but it's expensive and takes a long time. And sometimes businesses find themselves in situations where they need to take plan 'B' or even 'C'. You may have an MBA, but you seem to me to be a bit out of touch with the realities of technological businesses.
--

I have a dual degree in law and economic development. I still have questions about how a company would proceed given your analysis. At some point it had to be clear that what I will call the original business plan was not going to work. Joseph seems satisfied with your claims about chip yields; but to me that begs the question.

If your claims are correct (and they seem to be from my perspective) someone at Sigma should early on have known it was risky to expect good chip yields given the sensor specs. The design was, at best, pushing the envelope and at worst just plain silly. Once chip production began it became even more obvious things were not working as expected.

I am curious, if you had access to the foundry's design rules and how different the Sigma design was and as you speculate changing to a different foundry which prevented yield profiling plus other problems with the sd1 project would you consider delaying or even canceling the sd1.

God knows I have no problem taking risks, as anyone who has seen my pix will confirm, but at some point swimming against the tide becomes pointless. Personally I would have never let things get this far.
--
Those who forget history are condemned to go to summer school.

TomFlawrote:

bobn2wrote:

TomFlawrote:

bobn2wrote:

TomFlawrote:

SNIP

If Sigma really did proceed with a business plan so weak that chip price increases were as large as bobn2 claims, at least to me, that says a lot more about the business side of sigma than it does about the technical side.

I agree, to a point, but they were trying a highly innovative product, and innovation carries risks. My speculation as to what happened is almost all about the business side, not the technical side. One can imagine the sequence of events

SNIP

In any case to proceed knowing there was not a reliable chip source at a reasonable price defies any logic, no MBA would allow this to happen.

MBA or not, you are prognosticating about a business you know nothing of. Foundries sell a service. 'Reliable' means they deliver the semiconductor process accurately to the agreed specification. If the customer's design fails to work using that process, it is the customer's liability. The low risk route is to design conservatively, according to the foundries design rules, but as I explained, there may well be reasons that the Foveon engineers would not have done this. The question is, when would the Foveon management come to understand that the yield was below expectations.

These long winded speculations about what may have happened ignore the reality that no business would proceed with a business plan under the postulated circ*mstances.

Many businesses have - yield problems are reasonably common. Even Intel, the world's largest semiconductor company has suffered them, and failed to produce chips to its expected specifications. Read up on the whole NetBurst fiasco. A 'business plan' does not produce working chips, relentless debugging and yield characterisation does, but it's expensive and takes a long time. And sometimes businesses find themselves in situations where they need to take plan 'B' or even 'C'. You may have an MBA, but you seem to me to be a bit out of touch with the realities of technological businesses.
--

I have a dual degree in law and economic development. I still have questions about how a company would proceed given your analysis. At some point it had to be clear that what I will call the original business plan was not going to work. Joseph seems satisfied with your claims about chip yields; but to me that begs the question.

If your claims are correct (and they seem to be from my perspective) someone at Sigma should early on have known it was risky to expect good chip yields given the sensor specs. The design was, at best, pushing the envelope and at worst just plain silly. Once chip production began it became even more obvious things were not working as expected.

I am curious, if you had access to the foundry's design rules and how different the Sigma design was and as you speculate changing to a different foundry which prevented yield profiling plus other problems with the sd1 project would you consider delaying or even canceling the sd1.

God knows I have no problem taking risks, as anyone who has seen my pix will confirm, but at some point swimming against the tide becomes pointless. Personally I would have never let things get this far.

Companies have this particular problem, that they are run by human beings. Some of them have tried to eliminate this weakness, and employed MBA's instead, but somehow that never quite pans out either. Being run by human beings is quite risky, since while sometimes they produce incredible success, they also sometimes produce incredible failure. The failure of the SD1 is small beer as a compared with the world wide banking crisis as a story of foolish risk taking and failure to recognise that it was going pear shaped, but that still happened. There are many other similar examples. The problem with human beings is that they have tricky instincts like self preservation and they tend to go into denial when things aren't working out. This tends to mean that managements (not just MBA's) are not always working with full information and that sometimes technical folks pull the wool over their eyes about progress in pet products. That's easy to do, because the path to innovative development is rarely straight forward. management in technology businesses are used to the despair of 'nothing works' until suddenly (and magically to them) everything works. It's hard to predict. In the software industries managements have become quite used to this even in non-innovative projects.

So, in short, while one might not think that's what would happen in the best of all possible companies in the best of all possible worlds, I maintain that in real companies in the real world, that kind of thing is exactly what does happen.

That must be it - Sigma have invented a brilliant new marketing technique that will ensure their success.

Steve

amatiacwrote:

The ruse behind the scandalous price tag put upon the newly introduced flagship from Sigma house seems to be both simple and shallow one,

Minor companies like Sigma just need to dope their product marketing compaign with some spark of flamboyant ‘insolence’ in order to gain the podium of attention in the market dominated by sharks of Canon or Nikon caliber.

bobn2 wrote:

TomFlawrote:

SNIP

I am curious, if you had access to the foundry's design rules and how different the Sigma design was and as you speculate changing to a different foundry which prevented yield profiling plus other problems with the sd1 project would you consider delaying or even canceling the sd1.

SNIP

managements (not just MBA's) are not always working with full information and that sometimes technical folks pull the wool over their eyes about progress in pet products. That's easy to do, because SNIP

While I do not disagree with what you posted, in fact it seems like a reasonable explanation of what may have happened, a lawyer would object to your answer as non responsive.

I guess the bottom line is you seem to be arguing diminished capacity of some Sigma decision makers and possible mitigation due to misinformation from the techies; instead of answering the question "would you consider delaying or even canceling the sd1"?

Just as an aside thanks for sharing your technical insight about possible problems. If I got the basics of your posts there are precision problems when increasing the number of detectors on the sensor much faster than the physical size of the sensor size increases.

Thanks again for sharing your knowledge.
--
Those who forget history are condemned to go to summer school.

rf-design wrote:

I opt for "b2"

release with current bad yield and try to fix the issue and later slowly adapt >>the price to final target.

A yield issue should not necessary exist because of the three layered diffusion.

Not necessarily, but quite probably. The Foveon arrangement differs from the >standard three well CMOS in two important ways. 3 well CMOS implants N wells in >a P- substrate, and P- wells in an N- well in the P- substrate as you say. The >Foveon chip instead implants an N-well in the P-substrate, a P-well in the N-well. >Thus far it is the same as the standard process. Then it goes on to make a further >N-implant a in the P-well. That is an extra ion implant, and makes the process >different. (I know most CIS processes have extra implants, but my supposition is >that the one Foveon needs is not the run of the mill) Further, the depth of these >wells has to be precisely controlled, and as designed, since on the depth depends >the colour filtration properties. I have no idea how accurately the implant depth is >controlled in standard processes, simply because I've never had to know, but >those designing Foveon sensors do have to know, and to ensure that it is >accurate. Further, The Pixel design is complex. This is from their patent >US5965875,

That could be an argument. If the process is tested and modelled only for non-sensor mixed-signal application it is likely that a yield issue origin in the 3-diode 4 layer diffusion profile. From my practice I know that the depths are designed for voltage resistance and majority capacitance. Because the basic optical filter properties depend on layer thicknees there is a good reason to probably change this. For instance the leakage current is inverse to the diffusion gradient. But having a high gadient with changing polarity through the diffusion layers is diffucult. Furtheron the higher doping produce more vertical dislocation. So the typical weibull-distribution of the leakage would spread out more. So it is more likely that there are dead pixels. Is there are a technique to compensate dead pixels by interpolation if there are some?

For me there is no logical reason other than there exist a technical issue with has to be resolved. Beside the the critique I belive there are people who want to put the product out of the doors. I could explain to me that most of the considered unlogical reasons are because the anticipation of MURPHY and want it mean also to billion dollar companies is hard.

TomFlawrote:

bobn2 wrote:

TomFlawrote:

SNIP

I am curious, if you had access to the foundry's design rules and how different the Sigma design was and as you speculate changing to a different foundry which prevented yield profiling plus other problems with the sd1 project would you consider delaying or even canceling the sd1.

SNIP

managements (not just MBA's) are not always working with full information and that sometimes technical folks pull the wool over their eyes about progress in pet products. That's easy to do, because SNIP

While I do not disagree with what you posted, in fact it seems like a reasonable explanation of what may have happened, a lawyer would object to your answer as non responsive.

Sure, I forgot to make a specific response to that question.

I guess the bottom line is you seem to be arguing diminished capacity of some Sigma decision makers and possible mitigation due to misinformation from the techies; instead of answering the question "would you consider delaying or even canceling the sd1"?

I'm not arguing on their behalf, just making an observation on how things happen. there was a time in my career when I used to do 'due diligence' work for an investment company, going into companies they were considering investing in, and reporting back on the state of that company's technology base. The kind of situation I describe was amazingly common. There were dead-end developments that no-one in the company would drop, and they were pouring good money after bad. The decision as to when to cut and run from a no-hoper is one of the hardest for any company to make, but particularly smaller ones. Large companies pursue many technological developments and account in advance for the ones that will turn out to be DOA - that cost is simply loaded onto the successful ones. When you are a small company, as Sigma is in cameras, you simply can't afford to do that, which means there is much more percentage capital, both financial and personal, invested in the few developments. As to what I'd have advised had I gone into Sigma, I can't say, since I don't have access to the facts, but if my speculation is right then probably I'd have said:

• the SD-1 is never going to show a return, drop it.

• likewise Foveon. Clever idea, but not a good one. Time to stop.

• take stock of whether the body business really is important to the company as a whole (I suspect it's more aspirational, Sigma wanting to be a 'proper' camera company rather than just a third party lens supplier) -

If there is a case for staying in the - then play to the strengths of a smaller company, agility and ability to see and address niche markets (the game Cosina is playing rather well). Sigma has the SD-1 parts bin that can be targeted where they find easy pickings. Right now, perhaps

Drop the SA mount, adopt the EF mount (not hard, since the SA mount is the result of an illicit union between the K and EF mounts).
Buy the Sony IMX071 sensor
Pay attention to the details, such as the 1Mdot LCD from S-LCD

Load with all the 'pro' features that don't really cost anything (2 control wheels, 100% VF, metal body etc)
Sell at low/mid price (maybe $800 body).

OK, sales wouldn't be spectacular by Canon/Nikon standards, but I reckon that there would be a reasonable takeup from Canon users, if just to gain access to the Sony sensor without giving up their lenses, probably enough to make a margin, since development costs should be low.

Just as an aside thanks for sharing your technical insight about possible problems. If I got the basics of your posts there are precision problems when increasing the number of detectors on the sensor much faster than the physical size of the sensor size increases.

See Also

Updating the Model & Opening a Short-term Position in $TSM

I'll answer this in a separate post.

Thank you very much Bob for your insights and interpretation of the Sigma SD1/Foveon situation.

Hopefully, there will be some future solution that will ensure the viability of this technology for the masses.
--
S

Olympus E1 11-22 14-54 50 50-200 EC-14
Leica M4 28 35 50 90
Sigma DP1

I do not for one second think that this is Sigmas strategy.

dylanbarnhartwrote:

Honda loses $80K on each Acura NSX that it sells. Why bother selling it? Because it helps the sale of the Accord, the Civic, etc. People like to think Honda makes exotic cars.

Sigma is brilliant in making such a price announcement. It makes people compare Sigma brand to Leica, Hassy, etc. That has never been done before.

Let's face it, DSLR has never been an income generator for Sigma. Sigma makes money on lenses, not on cameras. Sigma DSLR will not sell in volumes regardless of price. Look at how cheap the last Sigma DSLR camera was without any success.

You are right, DSLR has never been a income generator. But Sigma top officials has stated that they want to be a player on the DSLR market, with a broader range of models, for their SA mount.

The high price made people wonder what about the SD1 that's so special. They eventually find out it's the Foveon sensor. People who do buy the Sigma DSLR's love them with passion. The Foveon sensor creates a cult-like behavior, much like the Leica's optics, the Hassy megapixel count, or the Alpa's mechanical precision. None of these brands sell in volumes, all of them are overpriced, and thus all of them are niche market products. It's fitting for Sigma to join the league.

Again, you are right about the cult thing. Sigma has slowly built a cult with their products, including the DP series that also has generated DSLR sales.

But, with the SD-1 affair, they have with one stroke managed to throw everything overboeard. Cult is almost too weak, family is closer. Sigma has built close relations with the SUG (Sigma User Group) with arrangments at trade shows, diners, use of voluntarly work force on stands etc. They have listen to our wishes, and used SUG people to beta test their cameraes. In all, a camera user community not found elsewere. With the announced price of the SD-1, Sigma changed it from a loyal fan base to a disappointed customer base. The issue is not the price in itself, but the fact that top Sigma Officials has stated the SD-1 would be in the 7D price range, and prior to launch has done nothing to tell that they would not be able to deliver at that price. It may be a cultural thing, but a existing/potentional customer in the western world will not put up with that. Sigma has alienated a lot of existing and potentional customers.

While nobody thinks Sigma's price is justified, most people believe the sensor is better than what Canon, Nikon, Sony put inside their professional DSLR's. That's no small feat.

Sigma's recent new strategy is clear: they want to move upscale. They don't want to be the cheaper, inferior quality vendor of OEM lenses anymore. They are generating a few lenses that match the OEM's best high-end lenses. They want to charge more money on these lenses, but people are hesitant. They need to make a strong statement that Sigma makes better products than the DSLR giants. They did that with the SD1.

I do not for one second buy that that was the intention. They ran into some unforseenable factors that made them change the price.

I think they invented the MF segment because of the price. The MF thing has not been mentioned before; SD-1 has been targeted as a prosumer DSLR camera.

Bravo, Sigma. I'm not buying your camera, but I'm looking into your new lenses.

rf-designwrote:

rf-design wrote:

I opt for "b2"

release with current bad yield and try to fix the issue and later slowly adapt >>the price to final target.

A yield issue should not necessary exist because of the three layered diffusion.

Not necessarily, but quite probably. The Foveon arrangement differs from the >standard three well CMOS in two important ways. 3 well CMOS implants N wells in >a P- substrate, and P- wells in an N- well in the P- substrate as you say. The >Foveon chip instead implants an N-well in the P-substrate, a P-well in the N-well. >Thus far it is the same as the standard process. Then it goes on to make a further >N-implant a in the P-well. That is an extra ion implant, and makes the process >different. (I know most CIS processes have extra implants, but my supposition is >that the one Foveon needs is not the run of the mill) Further, the depth of these >wells has to be precisely controlled, and as designed, since on the depth depends >the colour filtration properties. I have no idea how accurately the implant depth is >controlled in standard processes, simply because I've never had to know, but >those designing Foveon sensors do have to know, and to ensure that it is >accurate. Further, The Pixel design is complex. This is from their patent >US5965875,

That could be an argument. If the process is tested and modelled only for non-sensor mixed-signal application it is likely that a yield issue origin in the 3-diode 4 layer diffusion profile. From my practice I know that the depths are designed for voltage resistance and majority capacitance. Because the basic optical filter properties depend on layer thicknees there is a good reason to probably change this. For instance the leakage current is inverse to the diffusion gradient. But having a high gadient with changing polarity through the diffusion layers is diffucult. Furtheron the higher doping produce more vertical dislocation. So the typical weibull-distribution of the leakage would spread out more. So it is more likely that there are dead pixels. Is there are a technique to compensate dead pixels by interpolation if there are some?

Thank you for that. Your knowledge on this is clearly greater than mine, and it's nice to have my hunch corroborated as a possibility. I undertsnd that most commercial grade sensors have dead pixels, dead rows and dead columns, and are routinely interpolated. I suspect what might be a bigger problem for this sensor is non-uniform colour response. To get realistic colour from it requires non-linear colour mathematics, and I can imagine that if the colour response varied appreciably from pixel to pixel or area to area, one could get some pretty wild effects. Maybe there's a market in the Holga sector.

For me there is no logical reason other than there exist a technical issue with has to be resolved.

That's what I believe, too.

Beside the the critique I belive there are people who want to put the product out of the doors. I could explain to me that most of the considered unlogical reasons are because the anticipation of MURPHY and want it mean also to billion dollar companies is hard.

Many people fail to understand that Murphy has a seat on virtually every company management board.
--
Bob

s.humwrote:

Thank you very much Bob for your insights and interpretation of the Sigma SD1/Foveon situation.

You're welcome, so long as everyone understands that it s an interpretation, as you say, and not the facts.

Hopefully, there will be some future solution that will ensure the viability of this technology for the masses.

Frankly, I very much doubt it. If it is an expensive problem to sort out, I doubt that there is anyone who wants to put the money in, and the good old cooking Bayer seems to be more capable of delivering what the masses want.

As I said somewhere else, I think that the Foveon concept was a beguilingly clever idea, but in the end, not a good one.

(interestingly, there is a clue to why in the name. 'Foveon' is derived from 'Fovea' but the fovea operates more like a Bayer than it does like a Foveon - evolution has a way of finding good solutions, and in this case it's having spatially discrete colour receptors and interpolating in 'software')
--
Bob

TomFlawrote:

bobn2 wrote:

Just as an aside thanks for sharing your technical insight about possible problems. If I got the basics of your posts there are precision problems when increasing the number of detectors on the sensor much faster than the physical size of the sensor size increases.

That is the case, within certain constraints.

A semiconductor fabrication line is a very expensive asset. When a line is established it is designed to operated down to some minimum process geometry. The photolithography suppliers constantly work to reduce process limits, and that means that by and large new lines have the smallest geometries. These are usually dedicated to the strategic high volume digital products, processors, memory and the like - and there is a trade in second-hand lines, when a line is no longer competitive for memory, it will be turned to things like image sensors (with some adaptation of the processes).

When it comes to semiconductor yields, generally they will remain high until the design begins to approach the limits of the fab line, then yields can fall quite fast. My argument about the Foveon design is that it would meet the process limits earlier than a conventional sensor of the same pixel size.

The limits are not absolutely hard, some processing steps are more demanding than others so the minimum geometry might be different for different processes on the same line. Moreover, you don't find where the precise limits are until you characterise the process, making chips with reducing geometries until you find the yield going south.

So, the outcome is, that so long as your design is far from the process limits, you can go on shrinking. Once you get to the limits, if you still want to shrink, you have two alternatives - one is to redesign, eliminating whatever piece of geometry it turns out is causing the yield problem. The second is to upgrade the line or use a new one. Which is practicable depends on your scale and business process. If you are a 'fabless' semiconductor company, then moving to a new line means simply changing your fab supplier (or, if you use a large foundry, moving to another of their lines). If you are a huge volume player, like Sony, it is worth continually investing in new lines, and scheduling production around several lines depending on the demands of the specific chips. The companies with a general problem are small volume outfits that have their own fab line, such as Canon. If you look at canon's device history, they originally shrank without adjusting the process, because they were far from process limits. When they went to 15MP sensors they outshopped a part of the process (what's called BEOL - back end of line). In 2008-9 they instaled a new line which has enabled the 18MP sensors (the 7D sensor looks to have been fabbed on the old line, and is not quite as good as other variants of the 18MP sensor).

As well as that, as pixels get very small, more design tricks are required to maintain the optical efficiency.

So, in answer to your question, it's not straightforward, but in general making pixels smaller requires investment, either continuous or in periodic chunks, depending on the business model.

Thanks again for sharing your knowledge.

My knowledge comes from people more knowledgeable than me who shared their knowledge.
--
Bob

NT
--
The Lightmagician
Sun is my eye
Winds my breaths
Sky my open Mind.
MTFBWY
http://www.lightmagical.com

bobn2wrote:

TomFlawrote:

bobn2wrote:

TomFlawrote:

bobn2wrote:

TomFlawrote:

SNIP

If Sigma really did proceed with a business plan so weak that chip price increases were as large as bobn2 claims, at least to me, that says a lot more about the business side of sigma than it does about the technical side.

I agree, to a point, but they were trying a highly innovative product, and innovation carries risks. My speculation as to what happened is almost all about the business side, not the technical side. One can imagine the sequence of events

SNIP

In any case to proceed knowing there was not a reliable chip source at a reasonable price defies any logic, no MBA would allow this to happen.

MBA or not, you are prognosticating about a business you know nothing of. Foundries sell a service. 'Reliable' means they deliver the semiconductor process accurately to the agreed specification. If the customer's design fails to work using that process, it is the customer's liability. The low risk route is to design conservatively, according to the foundries design rules, but as I explained, there may well be reasons that the Foveon engineers would not have done this. The question is, when would the Foveon management come to understand that the yield was below expectations.

I have a dual degree in law and economic development. I still have questions about how a company would proceed given your analysis. At some point it had to be clear that what I will call the original business plan was not going to work. Joseph seems satisfied with your claims about chip yields; but to me that begs the question.

If your claims are correct (and they seem to be from my perspective) someone at Sigma should early on have known it was risky to expect good chip yields given the sensor specs. The design was, at best, pushing the envelope and at worst just plain silly. Once chip production began it became even more obvious things were not working as expected.

So, in short, while one might not think that's what would happen in the best of all possible companies in the best of all possible worlds, I maintain that in real companies in the real world, that kind of thing is exactly what does happen.

I've seen things that most people simply would not believe.

Companies form small centers of concentrated power. Emperors arise. And when the emperor gets an idea into his head, no one dares tell the emperor that he has no clothes.

At a previous job, upper management had this "diversification" theme, and some of the things that people built empires around were amazing. The lab next to mine was burning through about $20M/year (my estimate, based on the stuff they were building and the number of people involved) on a home natural-gas powered turbo generator, to be marketed as a full time mains replacement. As far as I could tell, someone figured that there would be a backlash against the electric vehicle philosophy and decided to make the exact opposite, personal "fuel into power" systems.

In a family owned, Japanese company, I would not have wanted to be the one to have to tell the COO, the owner's son, that there wasn't a chance in heck of meeting the cost targets for that sensor and that his pet project was in trouble.

I can see people saying "you tell him. No, you tell him" for 12 solid months, while the situation just spirals out of control more each day.

bobn2wrote:

• the SD-1 is never going to show a return, drop it.

• likewise Foveon. Clever idea, but not a good one. Time to stop.

There is also the DP series to consider and the entire SA lens line. I think what they will try to do is to keep the SD1 as a unobtanium halo product until they can fix the problems and introduce a lower price sibling with just enough fewer features to make it different (aka D700 vs. D3) and quickly followed by a DP version.

Load with all the 'pro' features that don't really cost anything (2 control wheels, 100% VF, metal body etc)
Sell at low/mid price (maybe $800 body).

This was similar to the strategy back in the SA film camera days. But it's really tough playing in the current market with both low volumes and low prices w/o some truly distinguishing characteristics; otherwise Pentax would be doing a lot better than they are now.

Sigma's bread and butter used to be the alternative "kit" lenses (i.e. inexpensive wide-to-short tele zoom + 70-300mm zoom) that dealers could bundle in with a low to mid price body to and sell for discount vs. the OEM products. But the OEMs have woken up and squeezed the life out of this model. Modern OEM kit lenses are so good and so cheap there little room for substitutes. Not to mention the OEM practice of shipping kits first and plain bodies later. (Sigma also made a boo-boo packaging the SD9 with their old kit 24-70mm and 70-300mm lenses. These lenses seriously under-performed the sensor.)

So Sigma, like Cosina before it, has realized that you can't underprice the big makers anymore in high volume products so instead you have to go where they are not. For Cosina it's bodies and lenses that have little development cost but are not longer made by the majors (e.g. rangefinder bodies and manual focus prime lenses). Sigma lately has been trying to build lenses that are not made or have not been updated by Canon/Nikon. So you have exotics like the 12-24mm/8-16mm rectilinear zoom, 120-300mm f/2.8 and more recently fast USM AF 30mm, 50mm, and 85mm primes. (BTW, Nikon has woken up to this and so you see the new 35mm f/1.8 and 50mm f/1.8. Canon is still shipping the 20+ year designs.)

Anyway, back in 2002 they tried to do the same with DSLRs. Sigma beat Minolta, Sony, and Pentax to this market and with something that was more than a "me too" product. Sigma also beat everyone to the large sensor compact market. They've done OK on innovation, it's been the execution that has been so-so and the follow through has been seriously lacking. You are seeing a company trying to re-invent itself - but in a semi-conservative Japanese way.

What they haven't done that Cosina has is to produce bodies for someone else's mount. But then again, Cosina only does this in a niche where "someone else" costs 5x to 10x more and they can underprice even used models.

Nikon has announced the Z6III, its third-generation mid-range full-frame mirrorless camera. A new 24MP sensor brings speed to every part of the camera, and all the key features have been upgraded.

The newest version of Panasonic's Micro Four Thirds video-oriented flagship camera has arrived, and it includes features like internal ProRes RAW recording, 32-bit Float audio capture, phase-detect autofocus, and compatibility with Panasonic's Real-Time LUT system and Lumix Lab app.

The Lumix S9 is Panasonic's newest full-frame mirrorless camera. It allows users to create their own custom looks for out-of-camera colors and is the first full-frame Lumix camera aimed squarely at social media content creators.

The Sony a9 III is the world's first full-frame mirrorless camera to feature a global electronic shutter with simultaneous readout. After extensive testing of this 120 fps sports camera, to see what you gain (and, perhaps, lose).

Fujifilm has announced the X-T50, a mid-range 40MP APS-C mirrorless camera that gains image stabilization, subject recognition AF and a host of high-res video features.

What’s the best camera for around $2000? This price point gives you access to some of the most all-round capable cameras available. Excellent image quality, powerful autofocus and great looking video are the least you can expect. We've picked the models that really stand out.

What's the best camera for travel? Good travel cameras should be small, versatile, and offer good image quality. In this buying guide we've rounded-up several great cameras for travel and recommended the best.

If you want a compact camera that produces great quality photos without the hassle of changing lenses, there are plenty of choices available for every budget. Read on to find out which portable enthusiast compacts are our favorites.

'What's the best mirrorless camera?' We're glad you asked.

Above $2500 cameras tend to become increasingly specialized, making it difficult to select a 'best' option. We case our eye over the options costing more than $2500 but less than $4000, to find the best all-rounder.