Photon-Counting Technology

A true photon-counting device can detect a single photon without read noise that degrades the performance of analog readout amplifiers. An example of a silicon photon-counting device is the Geiger-mode avalanche photodiode (GMAPD). Single GMAPD devices have been in existence and exploited for decades. We were among the first to successfully build arrays of GMAPD devices, and then integrate these arrays with a complementary metal-oxide semiconductor (CMOS) readout circuit directly behind the GMAPD array. Our first implementation of this technology measured the time of arrival of each photon. Each pixel had a register that counted pulses from a clock signal distributed to the entire array. When a pixel detected a photon, the counter located in that pixel is stopped. Readout of the stored counter values from all the pixels then provides the relative time of arrival of the first detected photoelectron for each pixel of the array. This device was then used to create a flash LADAR (Laser Detection and Ranging) receiver, in which a single sub-nanosecond pulse of laser light illuminates a scene. The reflected light from each part of the scene will arrive at the detector at a different time depending on its distance, therefore forming a three-dimensional image (instead of the usual two-dimensional intensity image of conventional detectors).

An APD is a pn diode that has high electric fields near the junction when reverse biased.  A photoelectron will be accelerated by this field and initiate a chain of impact ionizations that generates secondary electron-hole pairs. Thus, this photodetector has gain; one photon can yield many electrons of photocurrent flowing in the circuit. When the APD is biased below the avalanche breakdown voltage, the average gain is finite, as shown in Figure 1; this is because once a chain of impact ionizations is initiated, electrons and holes are extracted faster than they can multiply, causing the current to decay and die out. In this mode of operation, known as linear mode, the APD is a photodetector with fairly high-bandwidth internal amplification and can be used to measure optical signals with analog intensity variations. The linear-mode APD adds noise to the signal, since the amount of internal amplification varies with each set of signal photons.

Figure 1. Operation of photodiode, and both linear-mode and Geiger-mode APDs.Figure 1. Multiplication gain vs. bias of photodiode and both linear-mode and Geiger-mode APDs.

When the APD is biased above the breakdown voltage, its behavior is qualitatively different. The field is so high that, when an impact ionization chain is initiated, the electrons and holes multiply faster than they can be extracted, leading to exponential growth of current on a time scale of tens of picoseconds, as shown in Figure 1. The current level quickly saturates due to device resistance and space charge effects. Thus the APD acts as a digital device, switching from an off state to a conducting state in response to the detection of as little as a single photon. This mode of operation is known as Geiger mode. The GMAPD detects a single photon event without adding any read noise to the detection process, and therefore is a true photon detector.

In order to make use of the Geiger-mode detector for imaging, the pixel circuit must be able to detect the turn-on of the APD and then turn the APD off so that it can be reset to its initial state and be ready to detect another photon. The turn-off function is referred to as quenching the APD. Many investigators have reported sensing and quenching/reset circuits for various applications requiring a single detector or a small array.  For imaging, however, the circuit must fit within the real estate of a pixel, and small pixels enable high resolution. Our solution to the problem was to use a simple all-digital circuit in which the APD functions as a simple logic element.

Figure 2. Pixel biasing circuit LADAR timing circuit.Figure 2. Pixel biasing circuit and LADAR timing circuit.

The scheme is to bias the p side of the APD at a negative voltage (–VA in Figure 2) slightly less than breakdown. The n side of the APD is connected to a 5 V logic circuit (+VC in Figure 2). The n side is initially set to a logic high state by turning on the Arm pull-up transistor; this biases the APD at several volts above breakdown. The Arm transistor is then turned off, leaving the APD in an open-circuit configuration. When a photon is detected, the resulting avalanche discharges the APD capacitance, debiasing the APD to near the breakdown voltage, as shown in Figure 3, at which point the APD turns off. An inverter (shown in Figure 2) senses the discharge as a logic transition. For LADAR the rest of the pixel circuit times the detection event by freezing counts of a global High Speed Clock in the pixel Counter; for intensity imaging, a different circuit can be designed to count events and rapidly reset the APD after each event.

Figure 3. Output signal from APD after discharging.Figure 3. Output signal from APD after discharging.

Lincoln Laboratory has fabricated arrays of silicon GMAPDs in its Microelectronics Laboratory and prototyped arrays of CMOS digital timing circuits using the MOSIS foundry service. By fabricating the APD and the CMOS circuit on separate pieces of silicon, each component can be optimized separately. The APD can have a high fill factor because it does not share pixel real estate with the circuitry, and the doping profile of the APD can be optimized for a given operational wavelength rather than being constrained by the foundry process.   

We have developed hybridization techniques for bonding the APD arrays and CMOS chips together and thinning off the APD substrate, allowing back-illuminated operation. In a process we call “bridge bonding,” the APD and CMOS layers are bonded together with epoxy, the APD is thinned, and electrical connection vias are etched between the detectors through the thin APD layer reaching down to the CMOS layer underneath. Then metal straps are evaporated to connect the APD layer to the CMOS layer. Using this technique, 32 × 32-pixel focal planes have been fabricated and incorporated into a number of single-photon-sensitive flash LADAR systems.


top of page