Tuesday, December 12, 2017

IEDM Papers Review

Semiconductor Engineering publishes Mark Lapedus review of IEDM 2017 Imaging Session papers:

- TSMC and EPFL presented "a paper on what they call the world’s first back-illuminated 3D-stacked, single-photon avalanche diode (SPAD) in 45nm CMOS technology.

The SPAD achieves a dark count rate of 55.4cps/μm2, a maximum photon detection probability of 31.8% at 600nm, over 5% in the 420-920nm wavelength range, and timing jitter of 107.7ps at 2.5V excess bias voltage at room temperature.

- Sony presented "a paper on a CMOS photon detector - a non-electron-multiplying CMOS image sensor photon detector. Based on a 90nm process, Sony’s CMOS photon detector features 15μm pitch active sensor pixels with a complete charge transfer and readout noise of 0.5 e- RMS.

The pixel circuit is a conventional 4T pixel. The pixels are arrayed, resulting in a high conversion gain of 132uV/e-, according to the paper. The photodiode is expanded to a size of 14.7μm x 13.1μm in a pixel with a pitch of 15μm, resulting in a physical fill factor of 76% without using back illumination. 4 pixels in a column are simultaneously accessed and read.


  1. Crazy definition of a SPAD (Single Photon Avalanche Diode) if the detection probability is 5 % ...

    1. What's that supposed to mean? 5% at 920nm is not too bad, neither is 31% at 600nm.

    2. I think one key point to note regarding spads is that NOT every photon counts. It's a misconception that a spad is super-sensitive, and the photon detection efficiency plots of (all) spads show this. It is true to say that a spad avalanche event can be caused by a photon strike and subsequent impact ionisation (as well as a thermal or tunneling generated 'noise' carrier - i.e. dark count). The marking in time of this photon arrival via a digital signal's edge in relation to a.n. other event is the unique property that a spad brings to the party.
      TThe signal processing behind the statistics of many, many such marking events is the power of the application, be it e.g. proximity detection or fluorescence lifetime microscopy.

    3. Albert Theuwissen - Harvest ImagingDecember 13, 2017 at 7:07 PM

      I do agree, no other device has the same time resolution as a SPAD. Unfortunately the overall detection probability remains relatively low.

    4. Few remarks:

      - Timing resolution is limited by SPAD timing jitter - 107ps in one of these papers. These days, it's not that high accuracy. One needs to average many photons to get a better timing jitter.

      - SPADs have limited performance under strong sunlight. Depending on the scene reflectivity, optics and other factors, each pixel can easily get few sunlight photons per nano-second. With every avalanche event, pixel shuts down for quenching for few tens of ns or more, missing all other events during this time. SiPMs relax this a bit, but at a price. So, sunlight is definitely a hard problem for SPADs.


All comments are moderated to avoid spam and personal attacks.