What are the design challenges when using SPAD sensors?

Single-photon avalanche diode (SPAD) sensors are emerging on the market and are being used in automotive, medical, and industrial applications requiring extremely high sensitivity, precise timing, and 3D imaging capabilities, often in low-light conditions. Design challenges include managing dead time, suppressing noise sources like afterpulsing and dark counts, mitigating optical crosstalk, and achieving high photon detection probability (PDP) and high resolution, particularly when integrating SPADs into large arrays.

SPADs employ an avalanche process to generate significant current from a single photon. That results in higher power consumption compared with conventional CMOS sensors. In return for higher power consumption, SPADS can provide superior signal-to-noise (SNR) performance and other benefits.

Biasing and electronics

A basic SPAD consists of a p-n junction biased above the breakdown voltage (V_B). Their ability to detect single photons is a result of a positive feedback operation that occurs when the electric field is high enough to reach a critical value (>3.105 V/cm) by overcoming V_B with an excess bias (V_EX) bias voltage (Figure 1).

Figure 1. Key SPAD specifications include the breakdown voltage (V_B) and the excess bias voltage (V_EX). (Image: *Frontiers in Physics*)

SPADs are bistable devices and are either in a quiescent state or in avalanche. The device must be held in its quiescent state, biased above breakdown, for over a millisecond, waiting for the avalanche current injection. As soon as a photon generates an electron-hole pair, the avalanche is initiated with a rise time of less than 1 nanosecond, reaching a mA or more of current, enough to be detected by the readout circuit.

The avalanche current can be sufficient to damage the device if not quenched quickly. The readout circuit must detect the leading edge of the avalanche, generate an output pulse, and quickly reduce the bias to V_B and restore the initial operating level. This readout circuit is sometimes referred to as a sensing, quenching, and charging circuit, or simply the quenching circuit.

Performance considerations

SPADs can produce false counts from thermal effects, especially at elevated temperatures and increases in V_EX. The resulting dark count rate (DCR) can be managed with a combination of optimization of V_EXand maintaining the correct operating temperature of the sensor.

While dark counts are random, uncorrelated noise pulses, afterpulsing is correlated noise in SPADs. Afterpulsing results when trapped carriers generate a false signal (afterpulse) after an avalanche event. With dark counts and afterpulsing, there is no photon being detected. Techniques like active quenching or increasing the deadtime after an avalanche event can be used to minimize afterpulsing.

Pile-up can also be a challenge with SPADs. When too many photons arrive, the SPAD can be overwhelmed, producing inaccurate counts. Synchronizing the SPADs’ deadtime with the laser illuminator, and pile-up post-processing algorithms can be used to minimize this problem.

Data processing can be another challenge with SPADs. A large array can produce huge amounts of data. That requires an efficient readout architecture and high-speed data processing to enable sufficient throughput under peak operating conditions.

Signals, noise, and image quality

SPAD imaging performance using laser illumination in automotive LIDAR and similar applications can be improved by increasing the laser output, effectively reducing the signal-to-background-noise ratio (SbNR), and increasing the sampling rate. That increases the overall power consumption.

As expected, a low-power solution with low SNR and relatively few samples produces the highest variance in the estimated position of an object (Figure 2). While a high sample rate with a high SNR consumes more power, it produces the lowest variance in the estimated position. Combinations of low SNR and many samples or high SNR with few samples produce results with medium variances and intermediate power consumption.

Figure 2. Examples of the impact of different SNRs (SbNR) and sample sizes on SPAD system performance. (Image: *Scientific Reports*)

First arrival differential counting

First arrival differential (FAD) counting is used to simplify SPAD applications where the exact travel time of photons is needed, but not the precise arrival time. Instead of precise time stamps and complex per-pixel timing circuits like time to digital converters (TDCs), FAD uses the arrival time of the first photon to encode differential time information about subsequent photon arrivals and support calculations related to intensity or depth.

FAD is often used in applications like high dynamic range (HDR) imaging and flash LIDAR. In these applications, a full array of SPAD pixels is used. Capturing precise time stamps for each individual pixel results in complex circuitry, higher cost, and potentially less precise results. FAD can overcome those limitations.

Summary

SPADs are a rapidly emerging technology for LIDAR and other imaging and ranging applications. While they can provide superior performance compared with legacy technologies, they also require designers to use new approaches to realize maximum benefit, especially in power-sensitive scenarios.