Saturday, February 23, 2019

Actlight DPD Article

Laser Focus World publishes Lausanne, Switzerland-based Actlight article "Dynamic photodiodes reach single-photon sensitivity at low voltages, with minimal noise" by Maxim Gureev (Actlight CTO) and Denis Sallin (Director of Engineering). Few quotes:

"ActLight’s dynamic photodiode (DPD) uses a new operating mode that converts light intensity into time.1-3 The applied voltage is not constant, but is switched from reverse to forward bias. Applied forward bias induces a large forward current that doesn’t start immediately, but after a time delay (see Fig. 1). The forward current magnitude is controlled only by the applied voltage, and its value does not depend on the light intensity. In contrast, the delay time (triggering time) is a function of the absorbed light power."


"With CMOS compatibility, a high-amplitude output signal, low noise, and low voltage operation, the DPD can also be tuned to reach single-photon sensitivity. This could dramatically enhance the performance of wearable devices and smartphones."


"ActLight developed an iToF method using its DPD that can be manufactured with a standard CMOS or CMOS Image Sensor (CIS) process. The low-voltage iToF range meter demonstrator allows measurement at distances up to 5 m in a wide range of lighting conditions (see Fig. 3). Furthermore, it has demonstrated full sun (>100 klux) background light immunity."

13 comments:

  1. I was initially quite skeptical about this device but now I am much less so. Essentially in a PIN structure, the I layer is initially fully depleted. The capacitance of a 1um x 1um area, by 1um thick silicon depletion region might be (ignoring parasitics) 0.1fF. The presence of a single photoelectron inside this region can change the potential by about q/2C or about 750uV. This could be enough to change the switching time of the device from reverse to forward bias. (According to the LFW article, it is enough). see https://doi.org/10.1063/1.4906488
    It is not clear to me why this single-pixel device is really better than a modern single-pixel SPAD, but perhaps it is voltage or DCR. Probably it is not cost.

    ReplyDelete
    Replies
    1. I should add that the single photoelectron has to be in or very near the I-layer at the time the bias is switched in order to detect it by the proposed method. This means that the photon flux has to be fairly high by single-photon detection perspectives to detect anything at all. There is almost no integration period involved, if I understand things correctly.

      Delete
    2. I think that for a N+P_P+ diode, during reverse biasing, the I-zone is fully depleted and filled with negative charge. There is a potential barrier between N+/P-, so the sudden bias switch can not push electrons into the I-region(P-) before these negative charge be compensated. If light generates holes-electrons, then the effective negative charge should be less and elecrons can be pushed inside more rapidely.
      So the detection should be related to the negative charge inside the I-region(P-). Since the doping level is low, this number is low so single electron can give a significant influence ...
      For example at 1e12 doping level, a 1umx1umx1um volume contains only 1 dopant atom !!!
      am I right ?

      Delete
    3. The depletion region space charge disappears in nanosecond time, once the reverse bias is turned off. The bipolar carrier injection in forward bias mode is what takes some time to start.

      Delete
    4. So waht makes this start time so long in dark ??

      Delete
    5. The bipolar carrier injection does that. It's quite complex process and I have never comprehend it. I can understand the starting point of this process with zero current and its final steady state point with large forward current. But how exactly it behaves during transition between these two points is too difficult for my mental simulation abilities. Probably, a device simulator can show this process in fine details and answer on your question.

      Delete
    6. Forward conduction by a diode is by diffusion. As such, I would expect the delay to be limited only by carrier diffusion from the highly doped region into the the intrinsic region. Under a flat electrical field this only takes ~100ps to cross 1um. When the electrical field is not flat (due to the photoelectron), this "pushes back" against the diffusion mechanism. This could explain why you go from ps to ms.

      However, the longest lifetime of electrons in lowly doped Si is about 1ms. But thermal diffusion means it will have traveled centimeters in that time! This makes it highly improbably that the photoelectron has not been captured by diffusion.

      Have a look at: http://www.ioffe.ru/SVA/NSM/Semicond/Si/Figs/1323.gif for minority lifetimes.

      Delete
    7. YangNI, you mentioned "I think that for a N+P_P+ diode, during reverse biasing, the I-zone is fully depleted and filled with negative charge.". However, I don't think there is any negative charge om the I-zone at all. The are no mobile carriers there initially (hence Intrinsic), so under reverse bias, there can't be any space charge build up. Also, have a look at http://www.skyworksinc.com/uploads/documents/200823A.pdf for the section "reverse to forward". The only delay is due to diffusion from the highly doped sides.

      To quote the paper:
      "Typical total switching time [reverse to forward] can be on the order of 2% to 10% of the specified diode lifetime and in general is much faster than switching in the other direction, from forward to reverse"
      Again looking at the IOFFE shown above, the lifetime is ms, so switching is on the order of us. Looking at some Infineon papers, https://www.infineon.com/dgdl/AN034.pdf?fileId=db3a304313d846880113de8d362503df, lifetimes on the order of ns are mentioned.

      This makes me very skeptical about the claim of ActLight.

      Delete
    8. This comment has been removed by the author.

      Delete
  2. "In darkness, triggering times in the 100 ms range can be reached. For high intensities, sub-nanosecond triggering times are possible." How can a PN junction take 100ms to turn on under forward voltage biasing ?

    ReplyDelete
    Replies
    1. Definitely a confusing statement, I agree. Also seems to contradict Fig 4 where triggering is quicker with fewer photons. There is also the confusing statement in the text related to Fig 4: "The statistical distribution (histogram) of triggering time for different values of bias voltage corresponding to the critical charge equal to 1, 2, 3, 5, and 7 electrons are in good agreement with theoretical distributions obtained for integrating devices where charge accumulation can be described as a Markov process." Yet, the figure says experimental data - but why not use photons instead of some equivalent bias? Anyway, the work seems still a bit unclear as presented in LFW, and a little more clear in the paper I cited above.

      Delete
    2. Unfortunately, I don't have an access to the Applied Physics Letters paper, but I can imagine that PIN diode with thick i-region can exhibit a slow current buildup, just based on the general considerations.

      Indeed, suppose we have a thick intrinsic part and turn on the forward bias voltage. At first, the electrons can't enter the i-region from n+ side because it would create a huge space charge and repel them back. Same is true for holes from p+ side.

      In order to get the large current flow, we need both electrons and holes to be present in i-region, so that their space charges compensate each other. Eventually, if the forward bias is left there for a long time, the PIN diode gets to that state, but it takes some time.

      So, given the i-region is thick enough, the 400ms might be possible. Do they say something about the temperature in this experiment?

      Delete
    3. Meant to be 100ms instead of 400ms

      Delete

All comments are moderated to avoid spam and personal attacks.