CMOS avalanche photodiodes for sub-ns time-resolved images
Detectors with high spatial and temporal resolution are a requirement for fluorescence lifetime imaging (FLIM), Raman spectroscopy, time-of-flight optical ranging, and diffused optical spectroscopy. For these applications, imaging usually involves scanning systems with time-resolved detectors, as in confocal microscopes and lidar (laser illumination) scanners. The only commercially available wide-field image sensors with sub-nanosecond resolution are intensified charge-coupled device (I-CCD) cameras, which can be gated with windows as short as 200 picoseconds. However, these devices are typically used only in research and military applications because of their high cost.
Submicron CMOS technologies have shown potential for integrating avalanche photodiodes (APDs), which use avalanche multiplication to enhance the current generated through the photoelectric effect. These devices demonstrate good quantum efficiency (the ratio of converted electrons to incident photons) and low noise. Integrated APDs, combined with the parallel processing capabilities offered by CMOS electronics, promise a new generation of high-sensitivity, time-resolved image sensors. There are already medium-resolution pixel arrays capable of sub-ns photon timestamp measurement,1 but we have yet to obtain high-resolution image sensors because of various technological and architectural problems. For example, pixel electronics must provide ns time-processing and signal-storage capabilities while maintaining a small size and high fill factor (the ratio of photo-sensitive area to the total pixel area). We propose two approaches to fulfill these requirements, using APDs both in Geiger operation (where the APD is biased at a voltage larger than its breakdown voltage and the avalanche gain is very high) and in sub-Geiger (or linear) operation, where the APD is biased below its breakdown voltage and the avalanche gain is lower.
The first approach uses APDs in the Geiger mode, which are also known as single-photon avalanche diodes (SPADs). These provide a digital output pulse for each detected photon, and therefore the bandwidth required to process and transfer timing information for each photon in a high-resolution image sensor is generally too large to be manageable. To address this issue we first processed timing information using time-gating and stored the detected photon number at the pixel level, as is done in I-CCDs.2 With the second approach, we used APDs in the linear mode, by modulating their avalanche gain to obtain phase-sensitive detection.3 In this way, the APD itself works as a large bandwidth time-domain processing device.
There are already several examples of CMOS APDs fabricated using deep-submicron processes.4 When operated in the Geiger mode, these devices can have a timing resolution of a few tens of picoseconds and very low noise, if properly designed and fabricated using detector-grade processes.5 In linear-mode operation, deep-submicron CMOS APDs have a bandwidth in the GHz range and a low excess-noise factor.6–8
Although the excellent timing resolution of CMOS SPADs is an appealing feature, it is hard to exploit in a compact pixel. To address this issue, we can obtain sub-ns gating with small-size pixel electronics and acceptable power consumption. The key to minimizing the size of the pixel electronics is to transform the digital signal to the analog domain inside the pixel, storing each photon detection event as a charge packet in an integrating capacitor.9 We have shown it is possible to design a compact pixel with nanosecond timing resolution and good fill factor, without losing the single-photon detection capability. We created a prototype 32×32-pixel image sensor integrated in a 0.35μm CMOS process.9 The pixel, shown in simplified schematic form in Figure 1(a), has 25μm pitch and 20.8% fill factor. The sensor can be gated with a minimum time window of 1.1ns at a repetition frequency up to 80MHz. Figure 2 shows an intensity and a FLIM image acquired with the prototype sensor.
We exploited the possibility of modulating the voltage applied to APDs in linear-mode operation for the design of high bandwidth phase-sensitive pixels, as shown in Figure 1(b). When the system detects a modulated light signal, avalanche gain modulation mixes the optical and electrical signals, providing a phase-dependent output current. A simple charge-integrating amplifier can perform charge-to-voltage conversion in this mode of operation. We demonstrated a 64 × 64 image sensor based on this principle, and applied it in a time-of-flight 3D imaging system.3 We measured a very large demodulation contrast, reaching 80%, at frequencies as high as 200MHz. Figure 3 shows an intensity image and a range image acquired with the sensor.
We have shown here the potential of integrated APDs in imaging applications, but successful exploitation of these ideas in mass-produced devices lies with silicon foundries, which must undertake a technological effort to optimize the devices. If this happens, APD-based image sensors, combined with available low-cost pulsed and modulated light sources, could enable development of a new generation of instruments for time-resolved imaging applications.
Our efforts for the near future will focus on two complementary approaches. For one, we will work at the device level, to improve the detector and pixel performance. For the other, we will demonstrate arrays with higher spatial resolution and additional electronics, providing increased on-chip functionality and easing system integration.
University of Trento
Lucio Pancheri received his PhD in Information and Communication Technologies in 2006, and then served as a research scientist at Fondazione Bruno Kessler, Italy, until 2012. Since then he he has been an assistant professor.