Jon Stern

Lives in United States Mountain View, CA, United States
Works as a Electronics Engineer
Joined on Apr 21, 2003

Comments

Total: 126, showing: 1 – 20
« First‹ Previous12345Next ›Last »
In reply to:

Lawson Raider: I hope this isn't what the Hero 6 will look like, it is twice the size of the Hero 5 and will be a bulky monster.

It's Fusion. GoPro's new spherical capture camera.

https://www.wired.com/story/gopro-fusion-360-hands-on/

Link | Posted on Aug 8, 2017 at 15:26 UTC
In reply to:

Alberto Tanikawa: A curved sensor would not reduce vignetting, or improve peripheral illumination from lenses. That is dictated by the lens design itself. Light fall-off towards the edges will be exactly the same whether on a flat sensor, or elevated by a few millimeters on a curved sensor. More modern lens designs mitigate the need for such sensors. Not to say they are useless, far from it. Matching a telescope to a specific curved sensor should make coma a thing of the past.

@Enginel, polarization is not easy to achieve at the pixel level. It is an area of research though.

It's difficult to do better than the standard Bayer pattern. Attempts to do so have rarely gone far.

The two green diagonal pixels with a red and blue ends up being a good approximation for the human photopic response. If you try to add more colors you have a lower spatial sampling frequency in red, green and blue. Which would result in more moire, not less.

Link | Posted on Jul 17, 2017 at 07:07 UTC
In reply to:

Alberto Tanikawa: A curved sensor would not reduce vignetting, or improve peripheral illumination from lenses. That is dictated by the lens design itself. Light fall-off towards the edges will be exactly the same whether on a flat sensor, or elevated by a few millimeters on a curved sensor. More modern lens designs mitigate the need for such sensors. Not to say they are useless, far from it. Matching a telescope to a specific curved sensor should make coma a thing of the past.

@Airydiscus, With microlens shift, the sensor roll off with CRA is mitigated. I won't tell you specific numbers as I'm covered by a NDA by my last employer (a CMOS image sensor manufacturer), but the sensor roll off (thanks to microlens shift) is a smaller effect than the lens RI.

Having worked in the mobile phone sensor and module business for 6 years, I can tell you that I'm intimately familiar with this stuff. I've seen more mobile lens specs than I care to remember, and I've built and characterized dozens of mobile modules.

As for the max CRA, that's now significantly above 30°. Before back-side illumination technology, it was capped at about 28°, but with BSI and deep trench isolation (which helps control crosstalk) CRAs as high as 34° are relatively common.

Link | Posted on Jul 16, 2017 at 02:00 UTC
In reply to:

Alberto Tanikawa: A curved sensor would not reduce vignetting, or improve peripheral illumination from lenses. That is dictated by the lens design itself. Light fall-off towards the edges will be exactly the same whether on a flat sensor, or elevated by a few millimeters on a curved sensor. More modern lens designs mitigate the need for such sensors. Not to say they are useless, far from it. Matching a telescope to a specific curved sensor should make coma a thing of the past.

@AiryDiscus, I'm not quite sure what you mean.

For a start vignetting isn't the same as RI. Vignetting really means something blocking light rays (whether chief, or marginal rays). Whereas, relative illumination is the characteristic intensity roll of inherent in a particular lens design and manufacturing process (for the latter, particularly the quality of the coatings).

People tend to interchange the terms, so I'm not sure if you're saying that cellphone lenses have 100% RI, or there's nothing in the module that causes vignetting.

If you're saying there's 100% RI, that's not correct. The shading is corrected in the ISP using a lens shading correction (LSC) block that applies digital gain to the image according to a per-module calibrated shading table.

You don't see the RI and sensor roll off because it's hidden. In low light it will show up as noisier corners.

Link | Posted on Jul 16, 2017 at 02:00 UTC
In reply to:

steelhead3: There must be drawbacks for mass producing curved sensors; Sony released a camera a few years ago with one and we haven't heard anything more.

@Enginel, details? I'm not sure how I can give more details on the absence of a curved sensor in the product.

If you are wondering why we tore it down, I was working for a rival image sensor manufacturer.

Link | Posted on Jul 16, 2017 at 01:44 UTC
In reply to:

Ebrahim Saadawi: There must be some huge burden/drawback to the curved sensor concept, otherwise Sony would already have at least a few cameras with curved sensors.

They even showed a fully functioning camera with a fully completed design, lens (fixed), LCD, processor, card, and even full image samples. Yet, no wors from it being on the market.

Why? That little fixed-lens P&S probably costs 5 grand or something, making selling an impossibility.

It's like say Global Shutter readout replacing mechanical shutters, we all know it's a technological breakthrough in camera making history to remove the shutter (and mirror) from cameras, but then you look at a drawback and go humm... It's better. It removes any speed limits, it disrupts the entire DSLR and Mirrorless design and speed limits, gives a near infinite photo accuation limits, BUT... let's wait until we iron the kinks out till we can mass produce without making a stepback in another area. This other area can be price and can be technical (lowlight performance in the GS case).

So I see curved sensors giving us tiny ultra sharp wide glass, small f/1.4 FF primes with edge to edge sharpness at kit lens prices, sizes and weight, high perfornance superzooms, very very sharp pocketable fixed-lens cameras, just not now. In the future. I hope I am there.

@Josh Leavitt, partner with Sony? No!

Semiconductors Components Industries is the legal name of ON Semiconductor. They bought Aptina (where I worked), a rival CMOS image sensor manufacturer to Sony.

Link | Posted on Jul 16, 2017 at 01:41 UTC
In reply to:

Ebrahim Saadawi: There must be some huge burden/drawback to the curved sensor concept, otherwise Sony would already have at least a few cameras with curved sensors.

They even showed a fully functioning camera with a fully completed design, lens (fixed), LCD, processor, card, and even full image samples. Yet, no wors from it being on the market.

Why? That little fixed-lens P&S probably costs 5 grand or something, making selling an impossibility.

It's like say Global Shutter readout replacing mechanical shutters, we all know it's a technological breakthrough in camera making history to remove the shutter (and mirror) from cameras, but then you look at a drawback and go humm... It's better. It removes any speed limits, it disrupts the entire DSLR and Mirrorless design and speed limits, gives a near infinite photo accuation limits, BUT... let's wait until we iron the kinks out till we can mass produce without making a stepback in another area. This other area can be price and can be technical (lowlight performance in the GS case).

So I see curved sensors giving us tiny ultra sharp wide glass, small f/1.4 FF primes with edge to edge sharpness at kit lens prices, sizes and weight, high perfornance superzooms, very very sharp pocketable fixed-lens cameras, just not now. In the future. I hope I am there.

@Josh Leavitt, wafer thinning is used for back-side illuminated sensors. In that case the sensor wafer is bonded to a mechanical support wafer, or in the case of new stacked die sensors, to an ASIC layer, and then thinned.

For visible light CMOS sensors the epi layer is taken down to about 2.5µm. The support wafer is then thinned to bring the whole stack down to 150µm or 200µm (in most cases). To created curved sensors, you really need to thin down to <10µm.

If you're curious about methods of forming curved sensors, take a look at my patent:

http://appft.uspto.gov/netacgi/nph-Parser?Sect1=PTO2&Sect2=HITOFF&p=1&u=%2Fnetahtml%2FPTO%2Fsearch-bool.html&r=1&f=G&l=50&co1=AND&d=PG01&s1=20160286102&OS=20160286102&RS=20160286102

Link | Posted on Jul 15, 2017 at 07:42 UTC
In reply to:

Alberto Tanikawa: A curved sensor would not reduce vignetting, or improve peripheral illumination from lenses. That is dictated by the lens design itself. Light fall-off towards the edges will be exactly the same whether on a flat sensor, or elevated by a few millimeters on a curved sensor. More modern lens designs mitigate the need for such sensors. Not to say they are useless, far from it. Matching a telescope to a specific curved sensor should make coma a thing of the past.

Alberto, you seem to be assuming that lenses always achieve the best case relative illumination (RI) set by the cos^4 law. This is not the case.

Maybe high-end SLR lenses get close, but for many applications, we fall far short. Mobile phone lenses are a prime example. Many of the modern mobile lenses (where the manufacturers have been driven hard to reduce TTL) have RI around 35%.

That's not the whole story, because even with microlens shift (I guess that's what you mean by "microcells") the sensor still has some fall off in sensitivity for the high CRAs used in mobile.

Link | Posted on Jul 15, 2017 at 07:33 UTC
In reply to:

steelhead3: There must be drawbacks for mass producing curved sensors; Sony released a camera a few years ago with one and we haven't heard anything more.

Sony did not release a camera with a curved sensor. There were reports that the DSC-KW1/KW11 "perfume bottle" camera used one, but when we tore one down at my last company, the sensor was flat.

Link | Posted on Jul 13, 2017 at 04:23 UTC
In reply to:

Ebrahim Saadawi: There must be some huge burden/drawback to the curved sensor concept, otherwise Sony would already have at least a few cameras with curved sensors.

They even showed a fully functioning camera with a fully completed design, lens (fixed), LCD, processor, card, and even full image samples. Yet, no wors from it being on the market.

Why? That little fixed-lens P&S probably costs 5 grand or something, making selling an impossibility.

It's like say Global Shutter readout replacing mechanical shutters, we all know it's a technological breakthrough in camera making history to remove the shutter (and mirror) from cameras, but then you look at a drawback and go humm... It's better. It removes any speed limits, it disrupts the entire DSLR and Mirrorless design and speed limits, gives a near infinite photo accuation limits, BUT... let's wait until we iron the kinks out till we can mass produce without making a stepback in another area. This other area can be price and can be technical (lowlight performance in the GS case).

So I see curved sensors giving us tiny ultra sharp wide glass, small f/1.4 FF primes with edge to edge sharpness at kit lens prices, sizes and weight, high perfornance superzooms, very very sharp pocketable fixed-lens cameras, just not now. In the future. I hope I am there.

Josh Leavitt, silicon becomes quite flexible when thinned down to below a few tens of microns thick. That kind of thinness is used today for stacked memories (like in micro SD cards)

HowaboutRAW, no curved lithography is required. The sensor is formed using normal planar processes and then thinned and curved.

Link | Posted on Jul 13, 2017 at 04:20 UTC
In reply to:

Josh SZ: It will definitely make sensors much more expensive. But if, say, $100 cost increase on the sensor can reduce lens cost by $200 and can reduce the weight and size by 30%, then it will be worth the development.

Not necessarily. Having worked on curved image sensors and being a co-author of a patent (http://appft.uspto.gov/netacgi/nph-Parser?Sect1=PTO2&Sect2=HITOFF&p=1&u=%2Fnetahtml%2FPTO%2Fsearch-bool.html&r=1&f=G&l=50&co1=AND&d=PG01&s1=20160286102&OS=20160286102&RS=20160286102) for methods for forming them, I can tell you that this could be scaled to quite low cost in high volume.

Link | Posted on Jul 13, 2017 at 04:17 UTC
In reply to:

Alberto Tanikawa: A curved sensor would not reduce vignetting, or improve peripheral illumination from lenses. That is dictated by the lens design itself. Light fall-off towards the edges will be exactly the same whether on a flat sensor, or elevated by a few millimeters on a curved sensor. More modern lens designs mitigate the need for such sensors. Not to say they are useless, far from it. Matching a telescope to a specific curved sensor should make coma a thing of the past.

A lens designed for a curved sensor can deliver higher RI (less shading) while holding the other specs constant, than a lens designed for a flat sensor.

Equally, higher corner MTF can be achieved while holding other factor constant.

In a lens design, significant resources go in to creating a flat field. This comes at the cost of other design parameters.

Link | Posted on Jul 13, 2017 at 04:02 UTC
On article Fantastic footage of fjords with the DJI Mavic Pro (53 comments in total)

Nasty jello effect going on there.

It also looks like the there was a lot of noise in the video that's then been smudged by the compression.

Beautiful scenery though!

Link | Posted on May 10, 2017 at 22:35 UTC as 30th comment
On article GoPro unveils Fusion, a 5.2K spherical VR camera (43 comments in total)
In reply to:

Franz Weber: I am feeling dizzy now

Don't look at it using a mobile browser (well, at least Safari on iOS). If you use a desktop browser you get the correct VR Youtube playback that allows you to pan around.

Link | Posted on Apr 20, 2017 at 23:41 UTC
On article GoPro unveils Fusion, a 5.2K spherical VR camera (43 comments in total)
In reply to:

User9362470513: Genuine question - what's the point? It doesn't look good, induces sea sickness and looks like a gimmick that quickly wears thin. Not sure who would want to make or watch this sort of distorted footage.

Don't look at it using a mobile browser (well, at least Safari on iOS). If you use a desktop browser you get the correct VR Youtube playback that allows you to pan around.

Link | Posted on Apr 20, 2017 at 23:40 UTC

We had the same debate at work (in our Optical Systems Group) a couple of months ago. We ended up using this as our reference:

https://www.youtube.com/watch?v=OR8HSHevQTM

Link | Posted on Mar 30, 2017 at 20:00 UTC as 111th comment
In reply to:

justmeMN: "I can say it use to take about 70 people to make a lens like that prior to automation, now we need about 6 or 7."

In the USA, there is political discussion about "bringing back" manufacturing jobs. Increasing automation is why it's not going to happen.

@Teila Day, oh no! This is not the aftermath of automation. This is just the beginning. AI and improving robots will supplant more and more jobs.

Professional drivers? Gone in 10-20 years.
Pilots? Gone in 30 years (only so slow because of a airline industry conservatism).
Farm laborers? Gradually replaced by robots and mostly gone in a couple of decades.

And the list could go on and on.

Link | Posted on Mar 21, 2017 at 20:11 UTC
In reply to:

Jon Stern: What's missing from this tour is the coating. You go straight from polishing to assembly (actually, I'd expect an annealing stage too).

Did they show you the coating chambers? I'm particularly interested in the nano-coating technology and how this is grown.

@Dr_Jon, thanks for those links (I'm Dr Jon too!).

This answers several questions I have. It also tells me that this technology is only going to work for relatively large (low SAG) elements due to being spin coated. That in turn tells me that the approach is not going to work for the types of lenses that I work with professionally.

Link | Posted on Mar 21, 2017 at 18:01 UTC
In reply to:

Jon Stern: Among the many questions I would have asked is, "What are there so few DO lenses?"

Yes, but those compromises seem to be acceptable for some customers and use cases. Is it that demand is weak, or that they are expensive to develop?

Link | Posted on Mar 21, 2017 at 05:00 UTC
In reply to:

justmeMN: "I can say it use to take about 70 people to make a lens like that prior to automation, now we need about 6 or 7."

In the USA, there is political discussion about "bringing back" manufacturing jobs. Increasing automation is why it's not going to happen.

@Teila Day, the study was from 1995 to 2015, I think. I understand what the anecdotal evidence suggest, and I myself have spent a lot of time in China working with manufacturers there.

However, a detailed studies show that automation is the main reason for US manufacturing jobs losses, even while US manufacturing output has actually INCREASED.

http://fortune.com/2016/11/08/china-automation-jobs/

Link | Posted on Mar 21, 2017 at 00:27 UTC
Total: 126, showing: 1 – 20
« First‹ Previous12345Next ›Last »