Friday, March 09, 2012

Fraunhofer Lateral Drift PD Used for ToF Imaging

Laser Focus World: Fraunhofer Institute for Microelectronic Circuits and Systems (IMS; Duiburg, Germany) lateral-drift-field photodiode (LDPD) achieves a complete charge transfer from the pixels into the readout node in just 30ns - quite an achievement for 40 pixel. The researches used LDPD to create 128 × 96 pixel ToF sensor and a human arm was easily imaged in 3D using the sensor within a standard camera setup in conjunction with a 905 nm pulsed source (with a pulse duration of 30 ns) operated at 10 kHz. Responsivity of the LDPD was 230 μV/W/m2 and dynamic range was about 60 dB. The sensor is made in 0.35um process. The pixel fill factor is 38%.

Update: Fraunhofer's Annual Report gives more information about the LDPD pixel and ToF sensor:

Fig. 1. LDPD ToF pixel cross-section

"The photodiode is divided in two main parts: a pinned surface one and a part which resembles a buried CCD cell, as it can be observed in Fig. 1. The pixels and the entire sensor have been fabricated in the 2P4M automotive certified 0.35 μm CMOS technology at the Fraunhofer IMS with the addition of an extra surface-pinned n-well yielding a non-uniform lateral doping profile, as shown in Fig. 1 (upper picture). The doping
concentration gradient of the extra n-well was chosen in such a way that it induces an intrinsic lateral drift field parallel to the Si-surface in the direction of the pixel readout node (x-axis in Fig. 1) as well as from the periphery of the n-well in the direction of the n-well centre (y-axis in Fig. 1).

The potential distribution within this intrinsic lateral drift-field photodiode (LDPD) n-well resembles a hopper leading the photogenerated charge directly to the assigned readout nodes. It remains fully depleted during operation, sandwiched between the substrate and a grounded p+ pinning layer on top of it (see Fig. 1). In this manner, the almost noiseless reset and readout operations of the photodetector are enabled.

A buried collection-gate (CG) is fabricated at the one end of the n-well, which remains biased at a certain voltage VCG. It induces an additional electrostatic potential maximum in the system and enables the proper and symmetrical distribution of the signal charge among the readout nodes. Each of the four transfer-gates (TX) plays two main roles:

1) it serves to create a potential barrier in the well to prevent the collected charge to be transferred into any of the three “floating” diffusions (FD) aimed at pixel readout or the so called “draining” diffusion (DD) permanently biased at a reset potential
2) to facilitate the transport of the photocharge into a desired FD or the DD.


  1. At 10 uV/e-, which is a guess, gives a QE of about 7%. Another important parameter not given is demodulation contrast, DC at 16 MHz (assuming square wave modulation with 50% duty cycle). These two parameters are critical for depth accuracy. Also, lateral field drift for this application has been shown by others such as Tubert et al. at 2009 IISW. So, nice picture but some real data would be helpful if they want to boast about this sensor.

  2. Vlad, nice of you to add more data just as I was commenting on lack of information! Still, quantitative numbers would be good.

  3. Actually, they do refer to Tubert's paper as a prior art - see Fraunhofer's annual report pp. 47-50.

  4. More information can be found in their ESSCIRC 2011 proceedings publication:

    A. Spickermann et al., "CMOS 3D image sensor based on pulse modulated time-of-flight principle and intrinsic lateral drift-field photodiode pixels," Proc. ESSCIRC 2011.

  5. thanks for the paper reference. Still, it drives me crazy when people don't publish basic things such as FD conversion gain (we can guess but why not publish it??) and in this case, performance metrics such as demodulation contrast or depth error (Accuracy/precision)? A simple table would be helpful in ALL publications!

  6. Eric,

    I agree, perhaps it's good to start with a formal requirement for IISW papers to be accepted. If each publication there is required to have such table (where relevant of course) it may be possible to set a trend...


All comments are moderated to avoid spam.