Tuesday, March 12, 2019

Espros on ToF Sensing Know-How and QE Comparison with Sony, Melexis

Espros March 2019 Newsletter (not on their site yet but should appear soon) shares its CEO Beat De Coi thoughts on ToF imaging education:

"Did you know that the implementation of a TOF or LiDAR system needs professional competence in nine engineering majors? To me, that's the reason that probably nine out of ten TOF projects crash. At least this is my gut-feeling.

Some of the projects fail right at the beginning, which is good since the investment was limited. Others fail after years of trial and error engineering. Either way it is difficult to accept given the frustration and unrealistic expectations.

Remark: The University of Applied Sciences HTW in Chur/Switzerland offers a Bachelor of Science in Photonics since 2016. An important part of the lectures is TOF theory, implementation and application.

The Newsletter also compares Espros ToF pixels with Sony and Melexis competition:

"Competitor imagers have a 10μm pixel pitch whereas our epc611, 635 and 660 imager feature a 20μm pixel. The sensitivity is proportional to the pixel area which is 400μm2 in an ESPROS TOF imager, whereas it is just 100μm2 in the competitor devices. Smaller pixel pitch results in higher spatial resolution but at the cost of sensitivity which is more important in real world applications. All in all, the achievable operating range with the same illumination power is significantly higher with ESPROS TOF imagers."


  1. Let's assume every pixel is decently filled, which is anyway a requirement for indirect ToF (let's say e.g. >1ke-), read noise becomes negligible and you can think of binning smaller pixels without significant penalty. Moreover, you can think of structurally aware binning/averaging techniques that provide better performance at boundaries. I don't buy their argument on keeping a 20 um pixel is better. It's too general. In the end it boils down to how many photons do you get per area and does the shot noise of that overcome the read noise to make spatial-temporal averaging reasonable. Their claim is marketing nonsense. If their technology would be so great, they wouldn't have time to organize time-of-flight workshops, but would be drowning in getting resources for production for a real volume customer...

    1. I don't think the read noise is negligible. So all your assumptions are not supported.

    2. Readout noise would dominate at longer distances when there is no sunlight or if there is a very narrow optical filter. EESPROS has very competitive quantum efficiencies. However, their pixel's bandwidth pales in comparison to competitors.
      The focus on QE is good. But splitting a pixel into 4 pixels would reduce the effective readout noise as the noise gets ameliorated over 4 samples instead of 1.

    3. You can't get a decent depth estimate without a significant amount of photons. Won't do free consultancy here but if you don't mess up your design, shot noise is far more relevant than read noise. If you say your read noise is not negligible you need a redesign...

    4. "But splitting a pixel into 4 pixels would reduce the effective readout noise as the noise gets ameliorated over 4 samples instead of 1".
      So what? noise reduces by sqrt(4), signal reduces by 4. You get less SNR for the same amount of light if you split a large pixel into 4 smaller ones.

    5. I guess some people should attend Albert's courses...

  2. http://image-sensors-world.blogspot.com/2019/03/melexis-announces-tof-sensor-that-uses.html

    As indicated, the dark noise is 97 electrons. It looks quite high???

    1. They spec a FWC of ~160ke-. Assuming 1 V swing the resulting cap yields ~63e-rms read noise. They claim DC of 20-500 ke-/s. Assuming TINT of 0.1 sec you'd get 70e-rms from the dark current and a total of ~95 e-rms. The number seem to add up.

  3. Indeed, as long as package reliability, size and system cost do not matter then large pixels have higher sensitivity, no question here. However, when it comes to providing a reliable and commercially viable solution, a sensor with smaller pixel and built-in pixel binning is maybe a more effective solution, giving to the user options to go for high resolution or high sensitivity with pixel binning.

    Gualtiero Bagnuoli
    Marketing Manager Optical Sensors


All comments are moderated to avoid spam and personal attacks.