dark current density for DVXplorer?

[Peter M.]

What is the nominal value for dark current density (room temperature) for the DVXplorer? From hardware specs previously posted on the iniVation website, dark current density for DAVIS346 was listed as 2.88 fA (3.8nA/cm^2). However, I cannot find a corresponding number for DVXplorer. I did find a paper by B. Son, et al. [2017 IEEE ISSCC] which appears to describe the new event-sensor used in DVXplorer, but this doesn't provide a number for dark current.

Also, the current iniVation hardware specs for DVXplorer list of value of 0.3 lux under "dynamic range" ... is this on-chip illuminance?  

Thank you,

Peter

[Tobi Delbruck (INI)]

Great questions Peter.

The measurements of DAVIS dark current are not 100% solid, since our minds were not focused on it as much as for the DVS128 chip in

Lichtsteiner, P., C. Posch, and T. Delbruck. 2008. “A 128x128 120dB 15us Latency Asynchronous Temporal Contrast Vision Sensor.” IEEE Journal of Solid-State Circuits 43 (2): 566–76. https://doi.org/10.1109/JSSC.2007.914337.

The journal papers for the DAVIS cameras are

    DAVIS240

Brandli, C., R. Berner, M. Yang, and S. C. Liu. 2014. “A 240× 180 130 Db 3 µs Latency Global Shutter Spatiotemporal Vision Sensor.” IEEE Journal of Solid-State Circuits. https://ieeexplore.ieee.org/abstract/document/6889103/.

    DAVIS346:

Nozaki, Y., and T. Delbruck. 2017. “Temperature and Parasitic Photocurrent Effects in Dynamic Vision Sensors.” IEEE Transactions on Electron Devices. https://ieeexplore.ieee.org/abstract/document/7962235/.

Taverni, Gemma, Diederik Paul Moeys, Chenghan Li, Celso Cavaco, Vasyl Motsnyi, David San Segundo Bello, and Tobi Delbruck. 2018. “Front and Back Illuminated Dynamic and Active Pixel Vision Sensors Comparison.” IEEE Transactions on Circuits and Systems II: Express Briefs 65 (5): 677–81. https://ieeexplore.ieee.org/document/8334288/?arnumber=8334288&source=authoralert.

Let me know and I can share some unpublished results from the DVS128 paper preparation about dark current measurements. On that chip we had access to a pin that summed up all the photodiode currents and mirrored them out, so we could get a direct estimate of the average PD dark current. We also confirmed this by using a slow sinusoidal light intensity stimulus and finding the chip illumination at which the number of events dropped to half. At this point, the light intensity equals the "dark light" caused by dark current. Guesstimating the QE of about 50% times the fill factor, we could then infer the dark current in e/s.

For DAVIS chips, one way to directly infer a dark current is to use the APS output. Just expose a frame in darkness for some known time and see what exposure you get. Using the conversion gain from DN to electrons, you can infer the dark current. But our measurements of conversion gain were a bit hard to get reliable numbers for. We did the best we could to get the published numbers.

However, this measurement is also not quite correct for the DVS output. The DAVIS APS readout adds extra transistors that add dark current that the DVS path doesn't see. I think the best way to infer DVS dark current is by the contrast reduction method, and knowing the absolute chip illumination and the spectral QE.  See the DAVIS346 papers for the latest experimental results along these lines.

We (by that meaning INI) have not measured the Samsung VGA DVS; it would be cool to hear what they achieved.

In general the standards for such measurements are going through the same establishment as happened for CIS. The same thing happened with that field where it was unclear if the min illumination figures were for the chip or for the scene.

Tobi

[Chenghan Li]

Hi Peter, adding on to what Tobi has answered, the sensor used in the DVXplorer is indeed most similar to the one described in the 2017 ISSCC paper by B. Son, et al. The photodiode dark signal of this sensor is not published by Samsung, so we (iniVation) do not have this answer either. However, because this sensor is fabricated in standard CIS process, I would expect the photodiode dark signal to be not so different from any other photodiodes in conventional APS pixels. Regarding the 3lux/0.3lux values listed under "dynamic range", these are scene illuminance values.

[Peter M.]

Tobi, Chenghan Li,

Thank you for the quick replies and detailed response!

Peter

[Peter M.]

Chenghan Li,

A follow-on question regarding dynamic range:

What values are assumed for optic (f/#, transmission) and scene (reflectance) in terms of translating the 0.3 lux (low-end dynamic range) from scene illuminance to on-chip illuminance?  (I believe these are all the parameters needed.)

Thanks,

Peter

[li3.che...@gmail.com]

Hi Peter, sorry I didn't see your follow-up question until now. The 0.3 lux number should be taken as an indicative number estimated based on typical optic specs (that's good for low light) and scene reflectance. So you may use for example f/1.4, >90% transmittance, and a scene reflectance between 20-80% to estimate the corresponding chip illumination. Hope this helps.