High-rate Photon Counting with Geiger-mode APDs Mark A. Itzler, Mark Entwistle, and Xudong Jiang Princeton Lightwave Inc., 2555 Route 130 South, Suite 1, Cranbury, NJ USA 08512 Abstract—We extend the gate repetition rate of matched delay line transient cancellation to achieve 50 MHz pho- ton counting with low afterpulsing and to study after- pulsing effects in InGaAs/InP Geiger-mode avalanche diodes using sub-nanosecond gated operation. 1. Introduction The capability for high-rate detection of single photons in the near-infrared (NIR) wavelength range from 0.9 – 1.6 µm has become imperative for many applications in communications and optical sensing [1]. With greater interest in applying single photon detection to commu- nications-related fields such as quantum cryptography and free space laser communications, historical limita- tions on NIR photon counting rate in the range of 1 – 10 MHz have become a primary concern for many end us- ers. For low-light-level imaging, detection at the single- photon level provides the ultimate sensitivity possible, but low counting rates constrain image acquisition rates and dynamic range. The replacement of vacuum-tube based single photon detectors with solid-state alterna- tives is highly desirable but requires photon detection at high rates. In this paper, we describe some of the chal- lenges encountered in high-rate photon counting using InGaAs/InP Geiger-mode avalanche photodiodes (GmAPDs) and progress that has been achieved. 2. Challenges related to high-rate photon counting Geiger-mode operation entails biasing an APD above its breakdown voltage V b by an excess bias voltage V ex in a metastable state in which a single electrical carrier in- duced by a single photon can generate a macroscopic pulse of current that is easily sensed using electronic threshold detection. This avalanche current pulse must be quenched by lowering V ex sufficiently close to zero by either passive or active circuitry. One approach for realizing very high counting rates (up to GHz frequen- cies) is the use of high-speed switching of V ex on sub-ns time scales. The sub-ns rise and fall times associated with such short “bias gates” generate large capacitive transients that can couple to the output line when the gates are imposed on the SPAD. Therefore, fast-gating circuits require the suppression of these capacitive tran- sients to allow for the accurate avalanche detection. Beyond appropriate transient cancellation techniques, the main challenge to photon counting at high repetition rates is a rate-dependent increase in dark count rate known as afterpulsing. During each avalanche event, a small fraction of avalanche carriers can be become trapped at defect sites in the avalanche region of the GmAPD. At a later time, these trapped carriers are then released by thermionic emission. If the GmAPD is re- armed while trapped carriers are still being released, detrapping carriers can trigger additional dark counts. This “afterpulsing” effect can be avoided by introducing a sufficiently long hold-off time T ho to allow complete detrapping before re-arming the GmAPD, but the use of a long T ho constrains photon counting rates. The only recent efforts to effectively mitigate after- pulsing effects at high counting rates have focused on reducing the amount of charge trapped per avalance by minimizing the charge flow per avalanche. One method for reducing avalanche charge flow is to use extremely short excess bias gates, and this approach is consistent with high frequency operation when photon arrival times are sufficiently well-known. Any other technique used to rapidly quench avalanches following their detection would also limit charge flow (and subsequent charge trapping), and in addition to fast active quenching, the goal of rapid passive quenching has spawned interest in developing a variety of “self-quenching” diodes. 3. Transient cancellation by matched delay lines Bethune and Risk developed a transient cancellation scheme [2,3] based on the use of two matched delay lines―one inverting and one non-inverting―to linearly cancel the parasitic capacitive transients resulting from ns-scale gates, leaving any induced avalanche signal to be detected on a flat baseline. Past implementations of this scheme were limited to ~5 MHz operation [4,5] due to unacceptably high afterpulsing for gate repetition rates beyond this frequency. By improving the transient can- cellation of this approach and achieving reduction in current flow per avalanche, we have extended this tech- nique to demonstrate 50 MHz gate repetition rates with acceptable afterpulsing. The data in Fig. 1 illustrate cu- mulative afterpulsing probabilities (i.e., total afterpulsing probability per detected photon) as a function of photon detection efficiency (PDE) for frequencies between 1 and 50 MHz. The demonstration of 50 MHz operation with afterpulsing of 2.5% at 10.8% PDE represents a 5-10 X increase in gate frequency relative to past results with this technique. To obtain these data, we have de- veloped FPGA-based circuitry to count afterpulse events in as many as 128 bias gates following a primary photon- induced avalanche. This circuitry allows us to obtain data much more rapidly than using traditional time corre- 348 TuS1 (Invited) 2:00 PM – 2:30 PM 978-1-4244-8939-8/11/$26.00 ©2011 IEEE