ANGELO GULINATTI and NICK BERTONE
In the April 2005 issue of Laser Focus World, OptoElectronic Components (OEC) and colleagues presented developments from Micro Photon Devices (MPD; Bolzano, Italy) on a complementary metal-oxide semiconductor (CMOS)-compatible single-photon-counting detector based on a custom planar process. This type of detector offered picosecond timing resolution and low power consumption, making it very robust and well-suited for applications requiring fast timing resolution. However, when compared to the photon-counting detectors that use high-resistivity “thick” silicon and a dedicated process, planar photon-counting detectors have lower detection efficiency, especially in the near-infrared region.
While the “thick” reach-through structure provides very high detection efficiency, typically 75% at 650 nm, it also requires high bias voltage (400–600 V) and can only achieve timing resolution on the order of 300–800 ps. Alternatively, our planar structure of 2005 provided less than 30 ps timing resolution and needed very low bias voltage (making it very robust); however, the detection efficiency was a factor of 2.5 lower in the 650 nm range. Now, through the efforts of Sergio Cova of Politecnico di Milano, an early pioneer in the field of single-photon avalanche diodes (SPADs) and photon-counting electronics who has been studying these devices for the last 25 years, Cova and his team have designed a new planar structure that leads to the doubling of detector efficiency while keeping picosecond timing resolution, low dark counts, and low power consumption.1-3
Given a generic single-photon detector, it is possible to define photon detection efficiency (PDE) as the probability that a photon impinging onto the device active area is effectively detected. Once absorbed into the active region, the resulting photo-generated electron or hole must reach the high field region and trigger the avalanche.
Therefore, the PDE is a combination of the efficiencies of photon absorption and avalanche initiation. To gain a better insight into the detection process, a simplified one-dimensional representation of a planar detection structure can be visualized and four different regions can be identified: the substrate, the multiplying space charge region, and the two neutral regions at the two sides of the space charge layer (see Fig. 1).
Photons that succeed in entering the device are absorbed in one of the four layers, with a probability that depends on the absorption coefficient. If the photon is absorbed into the space charge region, a pair of photo-generated carriers are promptly separated and accelerated by the electric field. However, owing to the randomness of impact ionization processes, this does not guarantee that a self-sustained avalanche is triggered.
If a photon is absorbed into the lower neutral region (p-type), the photo-generated hole thermalizes with the other majority carriers, while the electron diffuses randomly through the region owing to the absence of a strong electric field. During its random walk it can recombine with a hole or it can reach the substrate. Since the electron is lost, in both the cases the avalanche is not triggered.
Conversely, if the electron reaches the multiplying space charge region, it is accelerated toward the high field zone and can trigger an avalanche. Therefore, the probability that the photon herein absorbed is detected is the product of the probability that the electron reaches the space charge region, the so-called electron collection efficiency, times the triggering efficiency.
A similar situation occurs if a photon is absorbed into the upper neutral region (n-type). The only difference is that now the minority carrier is a hole instead of an electron and that it will be lost if it recombines with an electron into the neutral region volume or at the interface with the silicon dioxide.
Finally, if the photon is absorbed into the substrate (n-type) it will be not detected since there is no way for the minority hole to initiate an avalanche in the active region.
A new planar processThe group at Politecnico di Milano has succeeded in developing a device that has high PDE at wavelengths greater than 550 nm—60% at 650 nm and 40% at 800 nm, while still achieving a timing resolution of less than 100 ps (see Fig. 2).
The starting point for this project was to develop a device structure with high PDE at wavelengths greater than 550 nm, while at the same time maintaining sub-100 ps timing resolution. An obvious approach to overcome this problem is to increase the overall thickness of the device active layer. In particular, it is possible to increase the thickness either of the space charge region or of the lower neutral region. The latter is certainly the easier solution to be implemented from a technological point of view since it does not influence the electric field profile inside the device and it only requires starting the process from a thicker epitaxial layer. However, this approach presents some drawbacks. Increasing the thickness of the neutral region results in a longer lifetime of the diffusion tail and a worsening of the timing resolution.
A more effective approach requires modification of the space charge region. But stretching out this region is more complicated than increasing the neutral layer thickness. In effect, the thickness of the depleted region cannot be increased by extending the low-doped region of the epitaxial wafer on which the device is fabricated. In this case, we would obtain a device with an electric field profile remarkably different from that of the 2005 devices. This would adversely affect all those device characteristics strongly influenced by the electric field profile, such as dark-count rate, temporal resolution, and avalanche triggering probability.
By changing the doping profile and modifying the starting material to obtain a multiplication field nearly identical to the 2005 devices, while at the same time increasing the thickness of the space charge region, we obtain a remarkable improvement in detection efficiency at long wavelengths. Since the dark count rate, avalanche probability, and temporal resolution mainly depend on the field in the multiplication region, this approach keeps them constant.
Andrea Giudice, CTO of MPD who has been working closely with the Milano group, says, “Although these new structures still do not have the detection efficiency of the reach-through structure, they are very close and given the much better timing resolution and robustness because of the low bias voltage, we are confident that they will be well received in the marketplace.” He also adds that these new devices will be incorporated into the photon-counting modules product line of MPD and that the group is continuing to explore new designs and processes to further increase the PDE.
From single pixels to a photon-counting camera
It is very important to understand the difference between a “thick” reach-through and a planar SPAD. The reach-through device requires a very custom process and high bias voltage (and it is nearly impossible to get an array of devices that will operate with the same bias voltage due to pixel-to-pixel variation and low yield). SPADs that are produced using a planar process have high yield and pixel-to-pixel uniformity, and are therefore well suited for production of monolithic arrays.For a 1 × 8 element SPAD array with each pixel having a 50 mm diameter and a 250 mm pitch, a microlens is incorporated to increase the fill factor (see Fig. 3). This 1 × 8 array can be used for parallel fluorescence lifetime measurements, and was developed for the European project ParaFluo (www.parafluo.com).
The ultimate goal for this technology is to produce a camera with single-photon sensitivity and picosecond timing resolution per pixel. Work on such a camera has already been demonstrated by using standard CMOS technology: the MegaFrame project (www.megaframe.eu) produced a 32 × 32 photon-counting camera using a standard 0.18 mm CMOS process. But while using this process allows for a high fill factor and the incorporation of the timing electronics, this low-voltage process also results in very poor PDE and small active-diameter pixels.
The group of Ivan Rech, who is leading the development of high-performance photon-counting detectors at Politecnico di Milano, envisions the development of a camera that uses the custom planar process to obtain large pixel sizes, high detection efficiencies, and picosecond timing resolution. In this case the detector pixels would be connected to the timing electronics through silicon vias and wafer bonding.
Rech says, “There is still a long way to go in producing a camera using our custom process; we will need to resolve issues like crosstalk and incorporation of vias to connect the SPAD to the electronics, but once these are solved, we strongly believe that this technology will give the best performance.”
1. A. Gulinatti et al., Proc. SPI 7355, 73550X, (2009); doi:10.1117/12.820661.
2. A. Gulinatti et al., J. Modern Opt., 58, 3, 210–224 (December 2010).
3. A. Gulinatti et al., Proc. SPIE 8033, 803302 (2011); doi:10.1117/12.883863.
Angelo Gulinatti is an assistant professor in the School of Engineering at Politecnico di Milano, Piazza L. da Vinci, 32 - 20133 Milano, Italy; www.english.polimi.it. Nick Bertone is president of OptoElectronic Components, 28 Des Lilas St., Kirkland, QC H9J 4A7 Canada; e-mail: [email protected]; www.optoecomponents.com.