39 megapixels vs. 4x5

On Sat, 21 Jan 2006 00:03:57 GMT, Mike wrote:


http://www.luminous-landscape.com/essays/Cramer.shtml

Doesn't rattle my world because I don't have $40,000 to dish out for a P45
digital back and I enjoy traditional B&W work with my view camera


Doesn't rattle it yet... but give it a few years.

I'm not sure if large chunks of processed
silicon will ever be "cheap" but maybe some
day they'll be "affordable."

Meanwhile, there are maybe one or two
outfits on the planet still making drum
scanners, and they're not cheap either.

An Epson 4990 is a lame substitute.
(Speaking from personal experience.)


rafe b
www.terrapinphoto.com

rafe b wrote:

I'm not sure if large chunks of processed
silicon

I'm alwasy so impressed with your grasp of
technical photographic and engineering
concepts...

On Fri, 20 Jan 2006 18:11:26 -0700, Tom Phillips wrote:



rafe b wrote:

I'm not sure if large chunks of processed
silicon

I'm alwasy so impressed with your grasp of
technical photographic and engineering
concepts...

* Silicon yield plummits as the die size increases. This has not changed
in 30+ years of silicon manufacturing. The largest microprocessor dies
are no greater than around 400mm^2.

* Current bayer sensors have inherent noise limits that will require a
huge technological breakthrough.

* Small sensors are diffraction limited

Given these three points, I believe that there will not be a cheap 4x5
film replacement anytime soon.

On Sat, 21 Jan 2006 01:24:26 GMT, Mike wrote:


* Silicon yield plummits as the die size increases. This has not changed
in 30+ years of silicon manufacturing. The largest microprocessor dies
are no greater than around 400mm^2.


Here's the great paradox. Semiconductor
technology has been pushing relentlessly
toward smaller feature sizes in order to get
ever-more active devices on "reasonably"
sized chunks of silicon. (400 mm^2 being
a very large chip, as you know)

But for photosensors, the need is for huge
areas of silicon -- and relatively huge features.
An "optimal" photosite with current CMOS
technology is around 5-7 microns on a side.

Compare this with current process feature
size limits, now at around 0.08 microns or less.
(0.13 um being fairly routine.)

So: a defect that kills a DRAM cell may have
little or no effect on a sensel that covers 25
square microns.

So it seems to me, the "classical" equations
for yield vs. die area may need some modification
for this particular technology.


* Current bayer sensors have inherent noise limits that will require a
huge technological breakthrough.


These alleged noise limits haven't prevented
DSLRs from utterly destroying the 35mm film and
camera market. Odd, that.

In fact, it's generally acknowledged tha DSLRs
are far *less* noisy than film, particularly at high
ISO values.

And no, I am not expecting cheap MF-sized
sensors any time soon, but that's more due
to economies of scale (or lack thereof) than
to technological limitations.

The mass market is already well served with
image sensors of 5x7mm. For a grand, you
get 15x22mm, which easily beats 35mm.

For three grand, you get 24x36mm, which
begins to encroach on 645 film territory

Those of us wanting MF or LF-quality digital
capture are left holding the bag. But it won't
take 6cm x 6cm of silicon to achieve that.

Even if you figure, say, $3000 for a processed
12 inch wafer, it's still hard to figure how these
MF backs command price tags of $40K.


rafe b
www.terrapinphoto.com

So: a defect that kills a DRAM cell may have
little or no effect on a sensel that covers 25
square microns.


But you are forgetting all of the logic and SRAM buffering that I imagine
is integrated with each sensel. I haven't studied the current CMOS
sensors, but I imagine there is a lot going on besides the capture of
light.

So it seems to me, the "classical" equations
for yield vs. die area may need some modification
for this particular technology.


Sure. I agree.


* Current bayer sensors have inherent noise limits that will require a
huge technological breakthrough.


These alleged noise limits haven't prevented
DSLRs from utterly destroying the 35mm film and
camera market. Odd, that.


Right. I think the low-hanging fruit has been claimed. But I'm skeptical
if we will see 15x22mm deliver 16-20 megapixels.


Those of us wanting MF or LF-quality digital
capture are left holding the bag. But it won't
take 6cm x 6cm of silicon to achieve that.


Right...I would guess 6x4.5cm of silicon will deliver 4x5" film quality
within 5 years.

Even if you figure, say, $3000 for a processed
12 inch wafer, it's still hard to figure how these
MF backs command price tags of $40K.


After you invest the $5 billion in the fab. Of course the niche players
will contract out the fab. I wonder who is manufacturing the PhaseOne
silicon?

Even if you find a fab, creating those masks is not cheap and existing
manufacturing processes are geared towards economies of scale (which you
correctly pointed out will not exist for MF/LF sensors).

Do you think a $3000 digital capture solution will be available in 5-8
years that can deliver 4x5 quality? I don't think so...but would value
your thoughts.

So: a defect that kills a DRAM cell may have
little or no effect on a sensel that covers 25
square microns.

So it seems to me, the "classical" equations
for yield vs. die area may need some modification
for this particular technology.


My other thought is that even when feature sizes where _huge_ (say 2-5
microns), they still couldn't get away with making large dies. Processes
have improved, but until wafers are manufactured in outer space, defects
happen.

rafe b wrote:

These alleged noise limits haven't prevented
DSLRs from utterly destroying the 35mm film and
camera market. Odd, that.

Mediocrity reigns...

On Sat, 21 Jan 2006 03:43:07 GMT, Mike wrote:

So: a defect that kills a DRAM cell may have
little or no effect on a sensel that covers 25
square microns.


But you are forgetting all of the logic and SRAM buffering that I imagine
is integrated with each sensel. I haven't studied the current CMOS
sensors, but I imagine there is a lot going on besides the capture of
light.


There's no SRAM, as far as I now. A linear CCD
is actually a pretty simple device -- an analog shift
register, really.

The only "digital" components in a linear CCD are
the clocks and transfer gates. The sensing
elements are simply capacitors.

I have experience and some expertise in linear
CCDs, but no experience in area arrays.

I'm guessing an area CCD would be treated as an
array of linear CCDs. I'm guessing there would be
analog multiplexers at the ends of the arrays, if only
to reduce the pin count and the number of analog
channels to some manageable level.


So it seems to me, the "classical" equations
for yield vs. die area may need some modification
for this particular technology.


Sure. I agree.


* Current bayer sensors have inherent noise limits that will require a
huge technological breakthrough.


These alleged noise limits haven't prevented
DSLRs from utterly destroying the 35mm film and
camera market. Odd, that.


Right. I think the low-hanging fruit has been claimed. But I'm skeptical
if we will see 15x22mm deliver 16-20 megapixels.


I'm skeptical also, unless the basic underlying
technology advances in some significant way.
Currently, the sensor in the D200 seems to represent
the state of the art for a 23.6 x 15.8 mm CCD.

With any given technology, you need to go to
a larger die to get more pixels of equivalent
quality. If you reduce the area of the sensing
elements, you get a more noise at any given
ISO rating.


Those of us wanting MF or LF-quality digital
capture are left holding the bag. But it won't
take 6cm x 6cm of silicon to achieve that.


Right...I would guess 6x4.5cm of silicon will deliver 4x5" film quality
within 5 years.

That would be nice. We can dream, can't we.

Even if you figure, say, $3000 for a processed
12 inch wafer, it's still hard to figure how these
MF backs command price tags of $40K.


After you invest the $5 billion in the fab. Of course the niche players
will contract out the fab. I wonder who is manufacturing the PhaseOne
silicon?

5 billion? Is it that high now? I've heard 1.5
billion, but that figure is a few years old.

Even if you find a fab, creating those masks is not cheap and existing
manufacturing processes are geared towards economies of scale (which you
correctly pointed out will not exist for MF/LF sensors).


Well, there are hundreds if not thousands of
so-called "fabless" semiconductor companies --
and I work for one of them. It's not too big a deal
nowadays. We've been averaging one major
new chip per year, and we started doing that
about ten years ago, with a team of only five
engineers, and a fairly small capital base.

I'm guessing a mask set and pilot run is on the
order of $250K to $750K, but that doesn't include
the initial engineering or layout, of course.

Standard Verilog or VHDL don't apply to
things like CCDs, so that's a diferent game
from the one I know.


Do you think a $3000 digital capture solution will be available in 5-8
years that can deliver 4x5 quality? I don't think so...but would value
your thoughts.

Your guess is as good as mine, Mike. The market
is a fickle thing. It's been a wild ride so far, hasn't it?


rafe b
www.terrapinphoto.com

rafe b wrote:
On Sat, 21 Jan 2006 03:43:07 GMT, Mike wrote:

But you are forgetting all of the logic and SRAM buffering that I imagine
is integrated with each sensel. I haven't studied the current CMOS
sensors, but I imagine there is a lot going on besides the capture of
light.

There's no SRAM, as far as I now. A linear CCD
is actually a pretty simple device -- an analog shift
register, really.
The only "digital" components in a linear CCD are
the clocks and transfer gates. The sensing
elements are simply capacitors.

Mike mentioned CMOS, which is somewhat different. But
you're right. Although the word "digital" is often used, the
imaging part of a CCD is an analog device, as is the readout
amplifier. That's why there's an A/D converter on board.

I have experience and some expertise in linear
CCDs, but no experience in area arrays.

I'm guessing an area CCD would be treated as an
array of linear CCDs. I'm guessing there would be
analog multiplexers at the ends of the arrays, if only
to reduce the pin count and the number of analog
channels to some manageable level.

There's a readout register. You shift a row into the
readout register, shift charge down the readout register
to the amplifier until you've finished the row, repeat.
There are one or a few amplifiers, not one per row.

Here are some pretty good pictures of how it works,
including a little animation of the readout process
(skip down to section 5):

http://www.astro.virginia.edu/class/...lec11-f03.html

One issue with really high pixel count detectors is
that the readout time through the small number of
amplifiers can become a bottleneck. Also, of course,
you do get cosmetic defects. Bad single pixels could
be interpolated over in software, since their location
is known, but a defect in the wrong place can cause a
bad column, which is pretty irritating but common in
research grade device (i.e. large, expensive, but not
NASA-level we'll pay through the nose for one without
defects expensive). Bad readout amplifiers ... now
that's bad news.

I'm skeptical also, unless the basic underlying
technology advances in some significant way.
Currently, the sensor in the D200 seems to represent
the state of the art for a 23.6 x 15.8 mm CCD.

With any given technology, you need to go to
a larger die to get more pixels of equivalent
quality. If you reduce the area of the sensing
elements, you get a more noise at any given
ISO rating.

Thermoelectric coolers, baby. Talk about your
battery hogs!

Even if you figure, say, $3000 for a processed
12 inch wafer, it's still hard to figure how these
MF backs command price tags of $40K.

My guess is that $3000 is an underestimate for
a CCD of that size and pixel count. (I'm talking about
the cost of just the CCD, not including the electronics
that control it.)

Mike wrote:
Do you think a $3000 digital capture solution will be available in 5-8
years that can deliver 4x5 quality? I don't think so...but would value
your thoughts.
If you are talking about an area sensor I see this as pretty unlikely,
mostly because the market is so small for such a camera. If you are
talking about a mechanically scanned linear sensor a $3000 camera is
very doable cost wise but not nearly as useful as a camera based on an
area CCD.

Scott