Open Access
Add to BibSonomy
Abstract
The overall sensitivity of frontside-illuminated, silicon single-photon avalanche diode (SPAD) arrays has often suffered from fill factor limitations. The fill factor loss can however be recovered by employing microlenses, whereby the challenges specific to SPAD arrays are represented by large pixel pitch (> 10 µm), low native fill factor (as low as ∼10%), and large size (up to 10 mm). In this work we report on the implementation of refractive microlenses by means of photoresist masters, used to fabricate molds for imprints of UV curable hybrid polymers deposited on SPAD arrays. Replications were successfully carried out for the first time, to the best of our knowledge, at wafer reticle level on different designs in the same technology and on single large SPAD arrays with very thin residual layers (∼10 µm), as needed for better efficiency at higher numerical aperture (NA > 0.25). In general, concentration factors within 15-20% of the simulation results were obtained for the smaller arrays (32×32 and 512×1), achieving for example an effective fill factor of 75.6-83.2% for a 28.5 µm pixel pitch with a native fill factor of 28%. A concentration factor up to 4.2 was measured on large 512×512 arrays with a pixel pitch of 16.38 µm and a native fill factor of 10.5%, whereas improved simulation tools could give a better estimate of the actual concentration factor. Spectral measurements were also carried out, resulting in good and uniform transmission in the visible and NIR.
© 2023 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
Single-photon avalanche diodes (SPADs) are reverse-biased diodes operating in Geiger mode, whereby a single photon can be detected with high timing resolution. SPADs perform direct photon-to-digital conversion, while read-out noise is virtually zero and dark counts can be low enough to enable Poisson-limited signal-to-noise ratios in photon detection. Nowadays, SPADs are available in most standard CMOS technologies, covering the entire spectrum, from NUV to NIR with peak photon detection probabilities (PDP) in excess of 50% in the visible spectrum, with dark count rates (DCR) of a few tens of counts-per-second (cps). Thanks to the economy of scale, the pitch of a SPAD can be as low as 2.2 µm, with an active diameter of 1.2 µm in a 4 × 4 test array, while a megapixel array size has been achieved [1,2]. Several architectures have been implemented over the years, spanning from linear to two-dimensional arrays on one hand, and from photon-counting to time-gating or time-correlated photon counting (TCSPC) on the other [3–5]. Several systems are available from commercial manufacturers and have been used for a number of applications in biophotonics, light detection and ranging (LIDAR), and true random number generation, just to name a few.
However, apart from a few notable exceptions [6], most frontside-illuminated (FSI) SPAD sensors are still characterized by relatively low fill factors (FF). Fill factor is defined as the ratio of the photosensitive area to the total pixel area; it is usually well below 50% in SPADs due to the area required by the high voltages used in SPADs, which in turn leads to limited photon detection efficiency (PDE). PDE is defined as the product of PDP and FF. This can be a problem for all truly photon-starved scenarios, as is the case in biophotonics. Researchers have used clever optical set-ups to employ the SPAD as a confocal pinhole, directly concentrating light on the photosensitive area, but this represents the exception rather than the rule and in any case it involves designs that are hard to build and/or to maintain [3]. In addition, most end users are accustomed to the high PDE of EMCCD and sCMOS charge-accumulating cameras, and often expect the same behavior from SPAD-based devices. Higher photon detection efficiency is therefore needed and can make the difference between a sensor that is adopted and becomes a product, and one that remains relegated to specific research experiments.
The main reason for the relatively low native FF of SPAD pixels implemented in FSI technologies is the need for in-pixel electronics – 10-transistor pixels are not uncommon, compared to the simpler 3T and 4T CMOS designs – as well as the need for a guard-ring structure to prevent premature edge breakdown [1,3]. This aspect is compounded by a relatively large pixel pitch, often in excess of 10 µm, which is closer to that of specialized CIS imagers, used e.g. in space [7], than that used in consumer or “scientific” products.
Microlenses can obviously be used to concentrate the incoming light to the SPAD active area, whereby high microlens efficiency over a wide angular distribution of light is often required. However, SPAD research prototypes and pre-series are often produced in small quantities, and include several designs or variants in a single silicon reticle, which is then reproduced on a full silicon wafer in a step-and-repeat procedure. The corresponding microlens multi-chip level replication presents an additional complexity, given that most microlens designs perform best when tuned to a specific pixel design and/or light distribution (e.g. low NA vs. high NA), whereas practical reasons dictate reducing the microlens replication steps per reticle. In addition, foundries addressing the consumer electronics market do usually target large to very large volumes, much larger than those of application-specific SPAD imagers.
In this work, we report on the design, simulation, fabrication and characterization of refractive microlens arrays by means of photoresist masters. The latter were used to fabricate molds for imprints with a UV curable hybrid polymer deposited on top of a family of front-side illuminated SPAD arrays. The main challenges we faced were represented by high pixel pitch (> 10 µm), small native fill factors (as low as 10%), and large formats (up to 10 mm). The overall goals were to ensure sufficient light collection for photon-starved applications, good uniformity in the visible spectrum, good spatial uniformity and high concentration factor (CF), defined as the ratio of output to input irradiance of the concentrator optical system [8], while moving to higher NAs and larger sensor formats when compared to the state of the art.
2. State of the art
Refractive microlenses placed on top of each pixel can be manufactured in several ways [9], e.g. resist processing (thermal reflow) followed by replication by thermal embossing or UV-casting from a master mold to a polymer resin [10–12], microdroplet inkjet printing [13] and other dispensing technologies [14], as well as MEMS-based or ultra-precision machining methods (see for example [15] with an emphasis on compound eye imaging systems). In general, it is important to carefully optimize all processing steps, including in situations where very dense microlens arrays are the target [16].
Pioneering SPAD-oriented work with refractive microlenses was described in [8], first at the theoretical level with the analysis of different non-imaging concentrators (e.g. conic, parabolic and tilted parabolic geometrical surfaces) and more traditional lens arrays. The former can in principle provide very high CF – even at high NA – albeit at the price of increased manufacturing complexity. Eventually, replica molded plano-convex polymer microlens arrays were fabricated [17], targeting the MEGAFRAME32 32 × 32 TCSPC imager [18], which featured 50 µm pixel pitch and a low native fill factor of 1%, i.e. the SPAD’s diameter was 6 µm. This was due to the in-pixel Time-to-Digital converter (TDC), memory, and read-out electronics. Batches of lens arrays were characterized on an optical bench before being mounted on the sensor, obtaining a CF up to 35 with good repeatability and back focal length (BFL). After a separate assembly consisting of alignment and gluing on top of the SPAD array, a CF of up to 25 was measured on a sizable area of the final assembled device, however with poor overall uniformity [19].
Refractive microlenses were also previously fabricated by the Authors, as from 2008-2009, using the process flow detailed in the next section, initially on the 128 × 128 “multi-sensor” time-resolved SPAD array manufactured in 0.35 µm CMOS technology and detailed in [20]. The corresponding modest median CF of up to 3, measured as the ratio of photon counts with and without the microlens array, was due to alignment issues and non-uniformity of the optical properties of microlenses. Process improvements were then applied to two other SPAD arrays in the same technology, namely the LASP 128 × 128 TCSPC array with column-level TDCs [11,21,22], and the SwissSPAD 512 × 128 binary gated imager [23]. Both detectors had similarly sized pixels with a pitch of 25 µm and 24 µm, respectively. The diameter of the active drawn region of SPADs was 6 µm and the native FF was 5%. The CF was measured to be 8-9 and 12 with a high level of collimation, i.e. with a lens of f/22 and f/10, respectively [24,25]. The lens sag was 4 µm and the optimal residual layer thickness (RLT, i.e. the height of the microlens base) for high f-numbers was around 45 µm, whereas, for low f-numbers a thinner residual layer led to higher CF. The CF uniformity becomes worse when increasing the f-number of the optical system, mainly due to the telecentric error [11]. Thickness variations and misalignments represent other possible error sources; in the case of SwissSPAD, for example, 14 different chips yielded an average RLT of 43.4 µm with 8.1 µm standard deviation, which is equivalent to an inter-chip CF variability of ∼25% at f/8.
Other refractive microlenses, manufactured using a similar approach, have been applied in [26] to a small 8 µm SPAD pixel in 130 nm CMOS technology. A cylindrical shape was selected in this case, to cope with the pixel arrangement in pairs of back-to-back rows, resulting in a fill factor of 50%, which is close to the maximum achievable for collimated light with a relatively large native fill factor of 26.8%. The RLT was equal to 14 µm, the sag to 2.5 µm, and a good photo response non-uniformity (PRNU) of 3.1% under uniform illumination over the 320 × 240 pixels were achieved when using optimal measurement conditions, i.e. an illumination tilt angle optimized to correct for microlens misalignments. A dedicated process was developed for SPADs in an industrial 40-nm CMOS technology [27], reaching a FF >70% at diode level and close to 40% at pixel level, whilst at the same time improving the overall PDP behavior by smoothing the very strong oscillations associated to the interferences in the optical stack. An average increase in PDP by a factor of 2 to 5% was also reported at 840 nm. Finally, refractive microlenses have been applied lately to backside-illuminated (BSI) industrialized SPADs fabricated in 45 nm / 65 nm 3D-stacked CMOS sensors [28], with a FF improvement from 31.3% to 50.6% for a 19.8 µm pixel pitch. Recently, other commercial processes have reported microlenses [29] on 10 µm square SPADs.
Approaches based on diffractive microlenses do typically employ binary-mask-based photolithography to generate a spatially-varying blazed grating structure, with rotational symmetry. They were proposed for example for CMOS imagers [30], as well as with SPADs, e.g. in [31] on 20 µm pixels, and used in [32] on the MiSPIA 32 × 32 sensor, having a native FF of 3.14% with pitch of 150 µm, and more recently with an enhanced design on the MF32 and MiSPIA 32 × 32 arrays with a pitch of 50 µm and 150 µm, respectively [33]. The designs reported in [32,33] employed fused silica substrates, which were then flip-chip bonded onto the sensors, resulting in good performance and homogeneity. Other diffractive approaches were used at wafer level on a 40-nm FSI SPAD fabricated in an industrial process employing amorphous silicon [34,35], while [36] proposed to integrate microlenses directly into the back-end of the SPADs by using a given metal layer to build Fresnel Zone Plates. The resulting microlenses have the advantage of being very compact and potentially amenable to integration in standard lithography steps. The resulting concentration factor is however usually wavelength-dependent, which can be an issue with applications requiring a broad spectrum. Angular acceptance is limited and the CF does usually reduce at low f numbers.
Future, disruptive approaches include the use of metalenses [37,38], for example those based on sub-wavelength structures, e.g. in the form of nanopillars [35,38,39], or holes milled in an appropriate substrate [40], with the potential for off-axis focusing in certain implementations [39]. One of the challenges with this approach consists in selecting (lossless) materials and processes compatible with CMOS fabrication while improving the wavelength and angular dependence of diffractive microlenses.
3. Materials and methods
3.1 SPAD sensors
The wafer reticle, i.e. the mask area which is reproduced on the whole silicon wafer multiple times in an optical lithography step-and-repeat approach, and onto which we imprinted microlenses in this work, contains a plurality of SPAD sensors with quite diverse features. The ones of interest to this work are the Piccolo [41,42], Ocelot [43,44], SwissSPAD2 [45,46] and LinoSPAD2 arrays [47,48].
Both Piccolo and Ocelot are SPAD image sensors with column-shared time-to-digital converters (TDCs), with 50 ps LSB and 12-bit range, placed at the bottom of the SPAD array, i.e. outside of the photosensitive area. Each column shares 4 TDCs in Piccolo and 6 in Ocelot to handle multiple SPAD firings in the same column in a short time. This architecture allows a native fill factor of 28% with a 28.5 µm pixel pitch. Ocelot-related measurements will actually not be reported in the following; however, given that the pixel is the same as Piccolo, the microlens design process and conclusions still apply.
SwissSPAD2 was, at the time of its publication, the largest SPAD array format ever reported. In this case, we selected a gated architecture with binary output and high read-out speed (up to 98 kfps) and reduced the pixel pitch to 16.38 µm (rounded to 16.4 µm from here on) in order to keep a manageable overall die size; this yielded a lower native fill factor of 10.5% for a round SPAD variant, and 13.0% for a square shape with rounded corners. LinoSPAD2 is a completely different, linear array with individual pixel raw read-out and a reconfigurable on-FPGA circuitry that can perform counting, time-stamping, and other functions. The architecture is similar to that of LinoSPAD with improved noise and sensitivity [49,50]. These arrays are presented in Fig. 1 and are summarized in Table 1.
Fig. 1. Micrographs of Piccolo, a 32 × 32 SPAD array with photosensitive area on the top section, highlighted in red (left) [41,42] – see also Fig. 8 (center); SwissSPAD2 512 × 512, a gated SPAD imager with 4 pixels shown in the inset (center, featuring round SPAD active areas in this case) [45]; Detail of LinoSPAD2, a 512 × 1 linear SPAD array with top alignment cross integrated in the metal stack (right).
Table 1. Main characteristics of the SPAD arrays which were studied in this work (measurements are available for all of them except Ocelot). All were manufactured in the same batch in 180 nm CMOS. All pixels are square while the SPADs have a round active area except for one SwissSPAD2 variant.
3.2 Microlens design
The main microlens design parameters, which we are going to use in the remainder of the paper, and which are related to the microlens fabrication process described in Section 3.4, are illustrated in Fig. 2. The lens radius of curvature, R, is linked to the sag, the diameter, D, and the conic constant, $\kappa $ ($\kappa = 0{\; }$corresponding to a spherical lens profile), as follows:
The microlenses designed in this work have a square footprint to maximize the fill factor of the square pixels on which they are placed. As their shape is obtained through a reflow process, they will exhibit a radius of curvature gradient between their edge center and corner, as depicted in Fig. 3 (left). This will yield a cross-shaped focal spot, visible in Fig. 4 (right). Our initial microlens masters were realized in quartz. As a consequence of the reflow and DRIE (Deep Reactive ion Etching) process we employed [9], the microlenses used in the early EPFL SPAD sensors up to 2014-2015 (see Section 2) had an aspherical profile, with $\kappa ={-} 1.6$. The following masters were however created with a thermal reflow process directly at CSEM, yielding a spherical profile. Therefore, $\kappa = 0$ was used in all the corresponding simulations, including the ones detailed in the following sections.
Fig. 2. Schematic microlens design parameters, here shown in cross-section. The RLT corresponds to the height of the microlens base.
Fig. 3. Simplified optical stack and optical system cross-section (left). It comprises from top to bottom: a 1.0 µm thick absorbing grid to model the surrounding interconnects/electronics, a 0.6 µm nitride passivation layer and a 6.9 µm intermediate layer with a constant refractive index of 1.5. Example of measured lens profiles exported to an STL file (right). The square microlens basis is clearly visible. The characterizations with optical 3D profilometers (Keyence VK-X1000 3D laser scanning confocal microscope) can however suffer from artifacts due to both reflectivity non-uniformity and large light angular incidence on the surface to be measured.
Fig. 4. Example of ray distribution when employing an f/22 objective, not shown in figure (left); tick marks are spaced by 5 µm on all axes. Light intensity distribution at the pixel surface when employing the Monte Carlo option [51] (right).
The sag is a key parameter, setting the microlens focal length through the radius of curvature (Eq. (1)), which does need to match the SPAD junction position in the silicon. It is basically fixed at reticle level, whereas the residual layer thickness can be changed from one imprint to another to reach the optimum performance for a given SPAD design. This has been exploited in our case, to cope with the design diversity.
3.3 Simulation tools and results
We have relied on two simulators for the microlens design for additional flexibility and to be able to cross-check our designs. They both calculate the trajectory and intensity losses of rays through the optical system. The first one is based on a commercially available package, Zemax OpticStudio in non-sequential mode, and it runs using a simplified version of the optical stack, as shown in Fig. 3 (left), placed between the SPAD and top sensor surface. It can operate on lens models or on 3D measured lenses by means of STL (stereolithography) files.
The second one builds on the in-house raytracing code detailed in [11,22,51], the main improvements being the possibility of importing a 3D measured lens via an STL file, see Fig. 3 (right), as well as the use of the full detailed optical stack of the CMOS process.
An objective can be placed in front of the SPAD; it is simulated by a thin lens whose sole purpose is to generate light rays hitting the system with incident angles corresponding to a given f-number (Fig. 4 left). Other illumination types include collimated light, with variable incidence angles given as parameters. A Monte Carlo option is also available, to randomly distribute the rays’ starting positions (and directions for the case of f-number simulation), shown in Fig. 4 (right). This allows the designer to obtain more accurate results using fewer simulated rays, thereby saving computational time, while at the same time reducing aliasing effects of the light distribution on the pixel surface, which may appear when using regular illumination patterns. The concentration factor in a given configuration is then calculated as the ratio of the number of rays crossing the SPAD active area with and without microlenses.
The optical simulations were used to drive the microlens design and fabrication. A uniform angular distribution was considered between normal incidence, 0°, and the maximal angle of incidence, AOImax. We considered a uniform 400–900 nm spectral distribution in order to calculate an average concentration factor (Eq. (3)), here defined from the Collection Efficiency, CE (Eq. (2)), as:
The product of the native fill factor, $F{F_{\textrm{native}}}$, and the concentration factor is also called effective FF, $F{F_{\textrm{effective}}}$ (Eq. (4)). The Collection Efficiency $CE$ takes into account the Fresnel losses in addition to the effective FF. Finally, the overall photon detection efficiency is equal to the effective fill factor multiplied by the photon detection probability (Eq. (5)), see also Section 1.
Figure 5 shows the obtained average CF plotted versus the lens sag and the RLT for an ideal lens with a gap of 1.6 µm, here for the SwissSPAD2-Round sensor. It is shown for two illumination schemes, one based on a low NA of 0.02 and a maximum AOI of ±1.15°, typical in microscopy, and one based on a high NA of 0.25 and a maximum AOI of ±14.48°, typical in near-infrared optical tomography (NIROT), LIDAR, or photography. Due to manufacturing constraints, the sag to be considered for SwissSPAD2 should be limited to a maximum of 6 µm and the RLT to a minimum of about 10 µm. From these results, a sag of 5 µm was therefore selected for the SwissSPAD2-Round sensor as a good compromise to cope with the two target NAs. Indeed, the ideal scenario would be to identify a common sag and residual layer thickness suitable for all sensors and for both NA cases.
Fig. 5. SwissSPAD2-Round simulated average concentration factors (ideal microlens profile) for a low NA of 0.02 (left) and a high NA of 0.25 (right). Insets: examples of ray distributions (96 rays, 5 µm sag, 10 µm RLT) using the same layout and color coding as in Fig. 3 left.
However, when using this 5 µm sag for the other sensors, namely SwissSPAD2-Square, LinoSPAD2 and Piccolo, Fig. 6 (left) shows that the optimal RLTs for the low NA case are quite different: below 20 µm for both SwissSPAD2 and 50–80 µm for LinoSPAD2 and Piccolo. In addition, Fig. 6 (right) shows that, for LinoSPAD2 and Piccolo, the optimal RLT range cannot be used anymore for the high NA case, where optimal results are obtained for an RLT of 25–45 µm. Neverthless, as LinoSPAD2 and Piccolo have a larger pixel pitch than SwissSPAD2, their microlens sag can be increased to 9 µm to get a single optimal RLT range (10–30 µm) valid for both NAs. Therefore, two microlens molds, one per sag, are required. A process allowing multiple sags on the same master could possibly overcome this limitation in the future.
Fig. 6. Simulated average CF for an ideal microlens profile as a function of the sag and RLT for a low NA of 0.02 (left) and a high NA of 0.18 and 0.25 (right). Different sensors are compared. The LinoSPAD2 NA value was chosen for compatibility with the measurement conditions in Section 4.4 (SwissSPAD2-R/S = SwissSPAD2-Round/Square).
3.4 Manufacturing methods (process flow)
Refractive microlenses for our frontside-illuminated silicon-based SPAD arrays were manufactured using an improved process flow based on UV-replication (or UV-imprinting) [7,11,52]. The original process flow relies on the origination of the positive microlens shapes to make masters (steps (a)–(c) in Fig. 7, herein on glass wafers) which are in turn used to manufacture the actual UV replication tool, called molds or stamps (steps (d)–(f) in Fig. 7). The latter consists of a negative of the microlens shape on a photomask. The same photoresist master can be reused to make other molds. The molds can be subsequently used to replicate microlenses on the production substrates, such as wafer, reticles, or bare dies, up to several hundreds of times. As we have seen, there are three critical microlens features: a small inter-lens gap (∼1.5 µm in this work) to reduce useless areas, the diameter (or side length for square-based microlenses), equal to the pixel pitch minus the inter-lens gap, and the sag. The first two parameters are defined in step (a), relying mostly on a photomask and photolithography process, while the last one is determined by the thermal reflow of the photoresist shown in steps (b)–(c).
Fig. 7. Main tooling and production phases of the microlens fabrication process flow in a microlens array (MLA).
Photolithography parameters, such as UV dose, recipe used in the development, and reflow temperature-time profile, were optimized to simultaneously obtain the microlens pitch and shape required to address the different imagers present on the same reticle. The final sag value depends mostly on the initial thickness of the photoresist pillars, obtained after step (a) and visible in step (b), but also on the microlens shape (round or square) and diameter. This is due to the volume difference of the photoresist pillars and the reflow mechanics.
We manufactured two masters to meet the previously detailed design specifications, including the target sags of 5 µm and 9 µm. Imprints were then carried out on a UV curable hybrid polymer deposited directly on top of the diced wafer reticles or individual dies, as shown in step (g) of Fig. 7. We initially employed OrmoComp (Micro Resist Technology GmbH, Berlin, Germany), but moved later on to BASF Lumogen OVD Varnish, whose lower viscosity allows addressing the challenging 10 µm residual layer target of SwissSPAD2 and which, if needed for quality reasons, can be delaminated to proceed with a new UV imprint on the same chip, as shown in step (j) of Fig. 7. The refractive indices at 589 nm are 1.520 and 1.508 for OrmoComp and Lumogen, respectively. The small difference can be compensated by finetuning the RLT. The Abbe number for both is 47, indicating weak chromatic dispersion.
The original process flow consists in aligning the mold to pixel arrays with µm-level precision and pressing on it to produce the desired microlens profile using a mask aligner (MA-6, Süss MicroTec, Garching, Germany) equipped with the UV imprint module, as shown in steps (g)–(i) of Fig. 7. This procedure was improved to imprint all chips of the reticle at once when using the 5-µm sag mold or only the two SwissSPAD2 dies when targeting the thinnest residual layer. Regardless of the imprint variant, the microlens material is UV cured and, after demolding, developed to remove the non-cured parts. In the case of OrmoComp, a thermal treatment completes the process. In the case of lower viscosity materials, there is an option to strip off the microlens array, if needed, to repeat the imprint cycle on the same substrate. This can be a welcome feature when targeting challenging imprint parameters. Finally, the reticles are ready for dicing and wire bonding.
The top microlens surface was not further processed against reflections. However, even if the top surface of the substrate does not have an anti-reflective coating (ARC) designed for the air interface, a microlens layer (even flat) can help to slightly reduce the parasitic reflections / Fresnel losses.
4. Measurement results and discussion
The measurement results have been divided in four sections, the first one (§4.1) dealing with the general microlens characterization procedure, followed by one section per main sensor type (§4.2-4.4). We employed different characterization set-ups, designed to address the main characteristics of each target sensor, as well as their key applications at the time of writing, for example, widefield microscopy for the large SwissSPAD2 array vs. NIROT for the Piccolo array and spectroscopy for LinoSPAD2. Table 2 provides a reference guide for the reader. All sections contain comparisons with simulation results.
Table 2. Measurement results guide for Sections 4.2–4.4 highlighting the main characteristics of the target SPAD arrays, the corresponding key microlens requirements, optical set-ups and key measurement results. Area = photosensitive surface, NIROT: near-infrared optical tomography, PRNU: photo response non-uniformity.
4.1 Microlens characterization procedure
The mold masters are optically inspected and their quality rated in terms of defects. A defect is classified as one or more missing or merged lenses, particles/residues, and residual stress on some lenses. The single 2 × 2 cm2 reticle layout is actually duplicated in the photomasks as a 3 × 3 array for redundancy, whereby any of such nine reticles, such as the one shown in Fig. 8 left, can then be used for the actual microlens replication on chip.
Fig. 8. Example of mold master quality control results after optical inspection (left). A single 2 × 2 cm2 reticle among those in the available 3 × 3 reticle array is shown. The different colors represent the quality control results, e.g. in terms of defects and non-uniformity (please see main text for details). The dies of interest are SwissSPAD2 (the two red squares at the bottom, in yellow), Piccolo (black rectangle at the bottom left, in green), and LinoSPAD2 (elongated blue structure in the center, in green). Microphotograph of the photosensitive area of the Piccolo 32 × 32 array with clearly visible alignment crosses, implemented using the top metal (center). Section of the LinoSPAD2 512 × 1 microlenses as measured with an optical 3D profilometer (right). Some measurement artifacts are clearly visible in the form of light surface undulations.
The actual sag and RLT of each microlens array were measured using a stylus profilometer (Tencor P-10, KLA Instruments, Milpitas, CA, USA) at the edge of each array prior to wire bonding and the results are reported in the next sections. It is interesting to note that, even when starting from the same photoresist thickness, the different pixel pitch, ranging from 16.4 µm to 28.5 µm, will induce slightly different microlens sag during the reflow step. The sag is generally lower for SwissSPAD2 than for Ocelot-Piccolo. This is due to the volume difference of the photoresist pillars and the reflow fluid mechanics. Microlenses deposited on Piccolo and LinoSPAD2 are shown in Fig. 8 (center) and (right), respectively.
UV-imprinted microlenses on SwissSPAD2 are shown in Fig. 9, illustrating the replication uniformity. The replication with most research and development mask aligners is particularly challenging when targeting the best possible performance parameters, due to the combination of the thin residual layer target (∼10 µm), large chip area, and pixel-to-microlens alignment precision tolerance of 1 µm or below, due to the actual SPAD diameter/width of ∼6 µm (see also Table 1).
Fig. 9. AFM scan (left) of single SwissSPAD2 microlenses, with ≈1.6 µm inter-lens gaps. Some artifacts are visible at the corners. This issue is being addressed in the latest fabrication runs (right), as clearly illustrated by the smoother optical 3D profilometry results and average inter-lens gaps of ≈1.0 µm.
4.2 Large area sensors (SwissSPAD2)
Preliminary microscopy-based SwissSPAD2 concentration factor measurements have been reported [12,46]. These measurements were followed by in-depth investigations with a benchtop optical set-up, cross-checked by means of the simulation environment detailed in Section 3.3.
The benchtop optical set-up allowed to measure the CF at any angle of incidence, to allow the assessment of the maximum CF. It consisted of a continuous wave collimated LED (M590L3-C, Thorlabs, USA) with a peak wavelength of 590 nm and bandwidth of 18 nm, which illuminated the sensors from a distance of 3.67 meters, guaranteeing uniform illumination of an entire 512 × 256 sub-array. We estimate that the pixel-to-pixel variation of the angle of incidence of the impinging light was smaller than 0.13 degrees, considering a sensor area of 8.4 × 4.2 mm2. The sensors were mounted on a PCB holder which allowed its rotation in two orthogonal directions. After an initial calibration step for each sample, in which the angle of the camera was optimized for highest photon sensitivity under fixed illumination settings while operating in live mode (real-time average intensity measurement), a sequence of intensity images was captured at this angle. The exposure time was carefully selected, so as to avoid high pile-up and to ensure sufficient SNR for both sensors. Data processing included pile-up correction, as described in [24], hot pixel removal (about 1% of the total number of pixels), and DCR correction by subtracting a reference dark image.
We used 7 SwissSPAD2 sensors with microlenses for this experiment, 5 with round SPADs and 2 with square SPADs with rounded corners, and 2 sensors without microlenses, one for each type of SPADs. The excess bias voltage (${V_{ex}}$) was varied as well, with a maximum of 6.5 V, which represents the preferred operation mode. The measurement results are shown in Fig. 10 and highlight significantly higher CF than the initial values (∼2.6) reported in [12,46], ranging from 3.65 to 4.2 for round SPADs, and from 3.2 to 3.6 for square SPADs with rounded corners. They are however still lower than the simulated performance for an ideal microlens, shown in Fig. 5 and in Fig. 6 (left).
Fig. 10. SwissSPAD2 average CF measured at various excess bias voltages. “R”: Round SPADs, “S”: square SPADs with rounded corners (see also inset). The results were obtained at the optimal angle of incidence for each sample, whereby the tuning was performed during the calibration stage before the actual measurements. Given that the fill factor of the square SPADs is slightly higher than for the round ones (13.0% vs. 10.5%), the resulting effective fill factor of the best samples is actually comparable.
It is interesting to note that the CF is largely unaffected by ${V_{ex}}$ from 4.5-6.5 V, while variations are more important in the 3.5-4.5 V range. This could be due to breakdown voltage differences between samples, whose importance is reduced in this SPAD design at higher values of ${V_{ex}}$ [53], leading to more uniform CF values.
The spatial distribution of the measured CF is shown in Fig. 11 for two representative samples. In each case we measured a 512 × 256 sub-array, whereby 16 × 16 spatial binning was applied for visualization purposes. Spatial gradients are visible, although these are somewhat emphasized by the tight color bar scale, the actual CF standard deviation of the respective mean CF values being 5.42% for sample E0R and 2.21% for sample D4S. CF measurements are actually known to display a larger relative spread at low NA, i.e. high f/# and resulting CF [11,24], such as employed in this measurement set-up, than at high NA. The light field non-uniformity was assessed by means of reference measurements carried out with a sensor without microlenses. The intensity standard deviation over all pixels was better than 1%, including the shot noise component. We therefore estimate that the illumination non-uniformity was better than 1%, and that the recorded CF values were indeed mostly due to microlens variability.
Fig. 11. SwissSPAD2 concentration factors: spatial distributions for two representative samples. In each case we measured a 512 × 256 sub-array. Round SPADs with rounded corners. Sample: E0R, CF std = 5.42% of mean CF value (left). Square SPAD with rounded corners. Sample: D4S, CF std = 2.21% of mean CF value (right). The results were obtained at the optimal angle of incidence for each sample, ${V_{ex}}$ = 6.5 V, spatial binning: 16 × 16.
The CF distribution measured for ${V_{ex}}$ = 6.5 V as a function of the actual RLT is shown in Fig. 12 (left). This is a crucial parameter for the SwissSPAD2 microlens performance; it follows the trend shown in Fig. 6 (left) for the ideal microlens case, highlighting a target residual layer thickness of ∼10 µm. The same figure also illustrates good control of the RLT variability. The measured sag vs. RLT distribution is detailed in Fig. 12 (center) for two batches, the second (“Batch2”, in red) being targeted at thin RLT for enhanced CF performance at high NA, as discussed in Section 3.3. Finally, Fig. 12 (right) reports the correlation of the mean measured CF over all ${V_{ex}}$ values, as reported in Fig. 10, as a comparison to enhanced CF simulation results. The latter were carried out by scanning a representative SwissSPAD2 lens with an optical profilometer and importing the corresponding STL file into the in-house raytracing simulator described in Section 3.3. About 105 rays, randomly initiated with the Monte Carlo option and with an f/# of 204 (basically corresponding to collimated light), were employed.
Fig. 12. SwissSPAD2 CF spread at the maximum excess bias ${V_{ex}}$ of 6.5 V (left). The ratio of the CF standard deviation (σ) to the mean CF value calculated over all pixels, also known as coefficient of variation [33], ranges from 1.48% to 5.42% (vertical error bars: ±σ). Distribution of sag values vs. residual layer thickness for the 7 SwissSPAD2 measured in Fig. 10 “Batch1” and 6 SwissSPAD2 of a more recent production run “Batch2” (center). Mean measured CF over all ${V_{ex}}$ values from Fig. 10 vs. simulated CF values, using the actual lens profile shown in Fig. 3 (right). The sag and RLT error bars are based on the peak-to-peak differences measured with a stylus profilometer at the chip edges and/or corners.
The difference between the measured and simulated results in Fig. 12 (right) is now much smaller than what was initially estimated and within the tolerances due to the measurement errors and lens shape variability. This confirms that the suboptimal concentration performance for SwissSPAD2 was mostly due to distortions of the lens shape, leading to optical aberrations. Conversely, the usefulness of tuning the camera angle in an initial calibration step points to a possible lateral shift in the focal point such as due to misalignment, similarly to what reported in [26]. The sensitivity to this effect, which was not noticed in the other sensors featuring larger pixels, can be worsened by the low initial fill factor and the resulting reduction in tolerances to misalignment errors.
4.3 Piccolo imager (32 × 32)
The test bench used for the concentration factor measurements of Piccolo included a supercontinuum light source (SuperK Extreme EXR-15 laser, NKT, Denmark) with an acousto-optic tunable filter, which was used to illuminate a fixed-focus collimator (FC/APC fiber collimator F220APC-850, Thorlabs, USA) placed in front of a movable diffusing element (2× crossed line diffusers, ED1-L4100, Thorlabs, USA), at a distance of about 80 cm from the sensor. The diffusing element, when in use, ensured uniform illumination intensity over the chip area. The laser was operated at 850 nm. The CF was measured by keeping the same optical setup and then switching the chips in and out, measuring the intensity over the whole array with a fixed integration time in each case, with and without the diffusing element. The actual CF was calculated as the ratio of the pixel intensities between chips with microlenses and a reference chip without microlenses, after background correction. When operating without diffuser, the total intensity over each full array was used, averaged over all pixels with valid output. When operating with the diffuser, a pixel-to-pixel comparison became possible after post-processing. The latter included artifact removal such as the elimination of hot pixels in all sensors and read-out artifacts (e.g. pixels registering very low counts). Pixels with missing values were replaced, wherever possible, with the median value of four neighboring pixels. Three columns with corrupted readings had their values replaced by the median of the adjacent four pixels (two to the left and two to the right).
The measurement results are summarized in Fig. 13 and Table 3. The best CF value with diffusers was 2.7 (2.97 without diffusers), allowing an increase of effective FF from 28% corresponding to the native FF to reach 75.6% and 83.2%, respectively, for this specific chip. Samples B2 and B4 have a sag value of ∼9 µm (vs. ∼5 µm for B1), close to the optimum, with a CF spread of only 1.7-2.7% of the mean CF value (standard deviation). Agreement with the values expected from simulation was in general within 15%. Overall homogeneity was also reported in samples B1 and B2, whereas B4 showed a slight SW-NE gradient.
Fig. 13. Piccolo CF after post-processing, as described in the text (diffuse illumination scenario). Chip B2 (left). Chip B4 (center). CF distribution for the three chips (right). White pixels correspond to missing measurements in the chips with microlenses and/or in the reference chip.
Table 3. Piccolo 32 × 32 imprint characterization and CF measurements results, compared with simulation results. The ratio of the total pixel intensities over the full array, averaged over all pixels with valid outputs, was used for the collimated light case, whereas a pixel-to-pixel comparison was employed with the diffuse light case.
4.4 LinoSPAD2 microlens spectral characterization
The test platform used for the spectral concentration factor characterization of the LinoSPAD2 512 × 1 linear array [47,48] is derived from the one described in [54], which is in turn based on the continuous light technique [55]. In detail, light from a wide-spectrum Xenon lamp is wavelength-filtered by means of a monochromator and fed to an integrating sphere to generate a temporally coherent and uniform spatial distribution. A calibrated photodiode (Hamamatsu S2281) with a large photosensitive area is connected to a precision source measurement unit (SMU) to act as reference. A light-tight box ensures the elimination of background noise sources. The LinoSPAD2 chip was placed at distance of about 14 cm from the output port of the integrating sphere (diameter of 3.7 cm, leading to NA∼0.177, and maximum AOI ±10.21°), to guarantee lower light levels and high uniformity [56]. This allows avoiding saturation and related pile-up effects, with a potential distortion of the SPAD’s sensitivity curve.
Figure 14 (left) reports the average CF of all pixels, calculated at each wavelength. Different bias voltages do not influence the average CF of 2.3, which is slightly below the simulated value of 2.9 for NA∼0.177. This might be due in part to imperfections in the lens shape and/or alignment. The error may also be due in part to the difficulty of precisely measuring the sag in this peculiar geometry. The transmission reduction in the NUV, below 400 nm, is due to the microlens material [57]. The pixel-to-pixel response, shown in Fig. 14 (right), features good uniformity, with a slight left-right gradient and a photon detection efficiency variation of 4.0% from one side of the array to the other (linear fit).
Fig. 14. LinoSPAD2 spectral characterization. Chip C2 is with microlenses, with a mean RLT/peak-to-peak variation of 19.85/0.9 µm, mean sag/peak-to-peak variation of 4.95/0.7 µm, ${V_{ex}}$= 4.5 V and Vdiode = 1.6 V. Reference chip NL6 without microlenses was measured with the same bias conditions. The CF at each wavelength is the mean value of the CFs measured over half of the array (256 pixels – contiguous for C2, alternating for NL6), reading out the chip as in [49,50] (left). Response uniformity at a fixed wavelength of 520 nm (right), for the same chip with microlenses (top curve), reference chip without microlenses (bottom curve), and CF (middle curve). In both plots the PDE and CF scales are shown.
5. Conclusions
Single-photon avalanche diodes (SPADs) are direct photon-to-digital detectors characterized by high temporal resolution and virtually zero read-out noise. Their overall sensitivity, when implemented in large frontside-illuminated arrays, has however often suffered from fill factor limitations due to the need for in-pixel electronics and guard rings to prevent premature edge breakdown. This can become a real obstacle to widespread use, in particular, when operating in photon-starved applications. The fill factor loss can be recovered by employing microlenses, whereby the challenges specific to SPAD arrays are represented by pixel pitch of 10 µm or higher, low native fill factor (as low as ∼10%), and up to 10 mm format. In addition, certain applications require a high light collection efficiency over a wide angular distribution, with a numerical aperture as high as 0.25.
Replications of refractive microlens designs were successfully carried out for the first time, to the best of our knowledge, at the wafer reticle level on different designs in the same technology in multi-chip operation, as well as on single large SPAD arrays with very low residual layer values (∼10 µm), as needed for better efficiency at higher numerical aperture. The smallest SPAD array employed was a 512 × 1-pixel line sensor, the largest a 512 × 512-pixel imager. Native fill factors ranged from 10.5% to 28%. Concentration factors within 15-20% of the simulation results were obtained for the smaller arrays (32 × 32 and 512 × 1), for example 2.7-2.97 for a 28.5 µm pixel pitch with a native fill factor of 28%, corresponding to an effective fill factor of 75.6-83.2%. CFs up to 66% of the simulation results were measured for the 512 × 512 imager after an initial calibration step to set the optimal angle of incidence. The corresponding discrepancy was well explained by an improved simulation tool when using the actual measured lens shape. Further performance improvement is expected, closer to the initial simulated concentration factor, as the microlens defects are being corrected and the inter-lens gap reduced to 1 µm, as shown in Fig. 9 (right). Spectral measurements were also carried out on the linear array, showing good and uniform transmission in the visible and NIR (no wavelength dependency), with some losses in the NUV due to absorption by the microlens polymer material itself.
In conclusion, we demonstrated that it is possible to achieve sufficient light collection for photon-starved applications, good uniformity in the visible, good spatial uniformity and high CF, while moving to higher NAs and larger sensor sizes with respect to previous work. The combination of low native fill factor and large chip format still calls for a high deposition accuracy and fabrication precision. These parameters improved SPAD imagers can address a large number of use cases.
Funding
Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung; Nederlandse Organisatie voor Wetenschappelijk Onderzoek.
Acknowledgments
This work was supported, in part, by the Swiss National Science Foundation (grants 20QT21_187716 Qu3D “Quantum 3D Imaging at high speed and high resolution” and 200021_166289) and the Netherlands Organization for Scientific Research (NWO) Project 13916. We would like to thank Guillaume Basset (CSEM SA), Angélique Luu- Dinh (formerly at CSEM SA, now at matchmycolor LLC), Christian Schneider (CSEM SA), Fabienne Herzog (formerly at CSEM SA) and Daniel Schlup (formerly at CSEM SA, now at Henkel AG & Co. KGaA) for the microlens tooling and production phase, as well as Benjamin Gallinet (CSEM SA) and Marc Schnieper (formerly at CSEM SA, now at SUSS MicroOptics SA) for previous collaborations. We would also like to thank Chao Zhang (formerly with EPFL, now at ADAPS Photonics Technology Co., Ltd) for the co-design of the SPAD arrays and imaging systems tested with microlenses, as well as Max Kollep (EPFL & Readily3D SA) for improvements to the EPFL-AQUA simulation code.
Disclosures
CB: PI Imaging Technology SA (I,S), IMA: PI Imaging Technology SA (I,E), EC: Fasttree3D SA (I,S) and PI Imaging Technology SA (I,S).
Data availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request
References
1. K. Morimoto and E. Charbon, “High fill-factor miniaturized SPAD arrays with a guard-ring-sharing technique,” Opt. Express 28(9), 13068–13080 (2020). [CrossRef]
2. K. Morimoto, A. Ardelean, M.-L. Wu, A. C. Ulku, I. M. Antolovic, C. Bruschini, and E. Charbon, “Megapixel time-gated SPAD image sensor for 2D and 3D imaging applications,” Optica 7(4), 346–354 (2020). [CrossRef]
3. C. Bruschini, H. Homulle, I. M. Antolovic, S. Burri, and E. Charbon, “Single-photon avalanche diode imagers in biophotonics: review and outlook,” Light: Sci. Appl. 8(1), 87 (2019). [CrossRef]
4. M. Caccia, L. Nardo, R. Santoro, and D. Schaffhauser, “Silicon photomultipliers and SPAD imagers in biophotonics: advances and perspectives,” Nucl. Instrum. Methods Phys. Res. Sect. A. 926, 101–117 (2019). [CrossRef]
5. I. Cusini, D. Berretta, E. Conca, A. Incoronato, F. Madonini, A. A. Maurina, C. Nonne, S. Riccardo, and F. Villa, “Historical Perspectives, State of Art and Research Trends of SPAD Arrays and Their Applications (Part II: SPAD Arrays),” Front. Phys. 10, 906671 (2022). [CrossRef]
6. I. Gyongy, N. Calder, A. Davies, N. A. W. Dutton, R. R. Duncan, C. Rickman, P. Dalgarno, and R. K. Henderson, “A 256 × 256, 100 kfps, 61% fill factor SPAD image sensor for time-resolved microscopy applications,” IEEE Trans. Electron Devices 65(2), 547–554 (2018). [CrossRef]
7. F. Zanella, G. Basset, C. Schneider, A. Luu-Dinh, S. Fricke, A. M. Madrigal, D. Van Aken, and M. Zahir, “Microlens testing on back-illuminated image sensors for space applications,” Appl. Opt. 59(12), 3636–3644 (2020). [CrossRef]
8. S. Donati, G. Martini, and M. Norgia, “Microconcentrators to recover fill factor in image photodetectors with pixel on-board processing circuits,” Opt. Express 15(26), 18066–18075 (2007). [CrossRef]
9. H. Ottevaere, R. Cox, H.-P. Herzig, T. Miyashita, K. Naessens, M. Taghizadeh, R. Völkel, H. J. Woo, and H. Thienpont, “Comparing glass and plastic refractive microlenses fabricated with different technologies,” J. Opt A: Pure Appl Opt. 8(7), S407–S429 (2006). [CrossRef]
10. P. Nussbaum, R. Volkel, H.-P. Herzig, M. Eisner, and S. Haselbeck, “Design, fabrication and testing of microlens arrays for sensors and microsystems,” EOS Pure Appl. Opt. 6(6), 617–636 (1997). [CrossRef]
11. J. Mata Pavia, M. Wolf, and E. Charbon, “Measurement and modeling of microlenses fabricated on single-photon avalanche diode arrays for fill factor recovery,” Opt. Express 22(4), 4202–4213 (2014). [CrossRef]
12. I. M. Antolovic, A. C. Ulku, E. Kizilkan, S. Lindner, F. Zanella, R. Ferrini, M. Schnieper, E. Charbon, and C. Bruschini, “Optical-stack optimization for improved SPAD photon detection efficiency,” Proc. SPIE 10926, 109262T (2019). [CrossRef]
13. J. Kim, N. Brauer, V. Fakhfouri, D. Boiko, E. Charbon, G. Grutzner, and J. Brugger, “Hybrid polymer microlens arrays with high numerical apertures fabricated using simple ink-jet printing technique,” Opt. Mater. Express 1(2), 259–269 (2011). [CrossRef]
14. P. Li, J. Pei, L.-F. Pan, and H. Zappe, “Fabrication method of low f-number microlens arrays by using surface coating and epoxy dispensing technology,” J. of Micro/Nanolith. MEMS MOEMS 13(2), 023004 (2014). [CrossRef]
15. W. Yuan, L.-H. Li, W.-B. Lee, and C. Y. Chan, “Fabrication of Microlens Array and Its Application: A Review,” Chin. J. Mech. Eng. 31(1), 16 (2018). [CrossRef]
16. S. Surdo, R. Carzino, A. Diaspro, and M. Duocastella, “Single-Shot Laser Additive Manufacturing of High Fill factor Microlens Arrays,” Adv. Optical Mater. 6(5), 1701190 (2018). [CrossRef]
17. S. Donati, E. Randone, M. Fathi, J. Lee, and E. Charbon, “Uniformity of Concentration Factor and Back Focal Length in Molded Polymer Microlens Arrays,” in Conference on Lasers and Electro-Optics, OSA/CLEO/QELS, JThE36 (2010).
18. M. Gersbach, Y. Maruyama, R. Trimananda, M. W. Fishburn, D. Stoppa, J. A. Richardson, R. Walker, R. K. Henderson, and E. Charbon, “A time-resolved, low-noise single-photon image sensor fabricated in deep-submicron CMOS technology,” IEEE J. Solid-State Circuits 47(6), 1394–1407 (2012). [CrossRef]
19. S. Donati, G. Martini, and E. Randone, “Improving photodetector performance by means of microoptics concentrators,” J. Lightwave Technol. 29(5), 661–665 (2011). [CrossRef]
20. Y. Maruyama, J. Blacksberg, and E. Charbon, “A time-resolved 128×128 SPAD camera for laser Raman spectroscopy,” Proc. SPIE 8374, 83740N (2012). [CrossRef]
21. C. Niclass, C. Favi, T. Kluter, M. Gersbach, and E. Charbon, “A 128×128 Single-Photon Image Sensor With Column-Level 10-Bit Time-to-Digital Converter Array,” IEEE J. Solid-State Circuits 43(12), 2977–2989 (2008). [CrossRef]
22. J. Mata Pavia, “Near-Infrared Optical Tomography with Single-Photon Avalanche Diode Image Sensors,” PhD Thesis #6481, EPFL, 2015. https://infoscience.epfl.ch/record/204779.
23. S. Burri, Y. Maruyama, X. Michalet, F. Regazzoni, C. Bruschini, and E. Charbon, “Architecture and applications of a high resolution gated SPAD image sensor,” Opt. Express 22(14), 17573–17589 (2014). [CrossRef]
24. I. M. Antolovic, S. Burri, C. Bruschini, R. Hoebe, and E. Charbon, “Nonuniformity Analysis of a 65-kpixel CMOS SPAD Imager,” IEEE Trans. Electron Devices 63(1), 57–64 (2016). [CrossRef]
25. I. M. Antolovic, S. Burri, R. A. Hoebe, Y. Maruyama, C. Bruschini, and E. Charbon, “Photon-Counting Arrays for Time-Resolved Imaging,” Sensors 16(7), 1005 (2016). [CrossRef]
26. I. Gyongy, A. Davies, B. Gallinet, N. A. W. Dutton, R. R. Duncan, C. Rickman, R. K. Henderson, and P. A. Dalgarno, “Cylindrical microlensing for enhanced collection efficiency of small pixel SPAD arrays in single-molecule localisation microscopy,” Opt. Express 26(3), 2280 (2018). [CrossRef]
27. S. Pellegrini, B. Rae, A. Pingault, D. Golanski, S. Jouan, C. Lapeyre, and B. Mamdy, “Industrialised SPAD in 40 nm technology,” in Proceedings of IEEE International Electron Devices Meeting (IEDM 2017), pp. 16.5.1–16.5.4.
28. A. R. Ximenes, P. Padmanabhan, M. Lee, Y. Yamashita, D. N. Yaung, and E. Charbon, “A 256×256 45/65 nm 3D-stacked SPAD-based direct TOF image sensor for LiDAR applications with optical polar modulation for up to 18.6 dB interference suppression,” in Proceeding of IEEE International Solid-State Circuits Conference (ISSCC 2018), pp. 96–98 (2018).
29. O. Kumagai, J. Ohmachi, M. Matsumura, et al., “A 189×600 Back-Illuminated Stacked SPAD Direct Time-of-Flight Depth Sensor for Automotive LiDAR Systems,” in Proceedings of IEEE International Solid-State Circuits Conference (ISSCC 2021), pp. 110–112.
30. V. Rochus, X. Rottenberg, I. De Wolf, P. Soussan, and P. De Moor, “Pixel performance enhancement by integrated diffractive optics,” in Proceedings of IISW, 82–85 (2015). https://www.imagesensors.org/Past%20Workshops/2015%20Workshop/2015%20Papers/Sessions/Session_4/4-04_ROCHUSJayapala.pdf.
31. S.-C. Cheng, C.-C. Chang, K.-F. Lin, C.-H. Huang, L.-Y. Tseng, H.-M. Yang, K. Wu, and J. C. Hsieh, “Lens Solution for Intensity Enhancement in Large-Pixel Single-Photon Avalanche Diode,” in Proceedings of IISW, R17, pp. 62–65 (2017). https://www.imagesensors.org/Past%20Workshops/2017%20Workshop/2017%20Papers/R17.pdf.
32. G. Intermite, A. McCarthy, R. E. Warburton, X. Ren, F. Villa, R. Lussana, A. J. Waddie, M. R. Taghizadeh, A. Tosi, F. Zappa, and G. S. Buller, “Fill factor improvement of Si CMOS single-photon avalanche diode detector arrays by integration of diffractive microlens arrays,” Opt. Express 23(26), 33777–33791 (2015). [CrossRef]
33. P. W. R. Connolly, X. Ren, A. McCarthy, H. Mai, F. Villa, A. J. Waddie, M. R. Taghizadeh, A. Tosi, F. Zappa, R. K. Henderson, and G. S. Buller, “High concentration factor diffractive microlenses integrated with CMOS single-photon avalanche diode detector arrays for fill factor improvement,” Appl. Opt. 59(14), 4488–4498 (2020). [CrossRef]
34. J. Vaillant, L. Masarotto, R. Paquet, V. Lecoutre, C. Pellé, N. Moussy, and S. Jouan, “SPAD array sensitivity improvement by diffractive microlens,” in Proceedings of IISW, P28 (2019). https://www.imagesensors.org/Past%20Workshops/2019%20Workshop/2019%20Papers/P28.pdf.
35. L. Dilhan, J. Vaillant, A. Ostrovsky, L. Masarotto, C. Pichard, and R. Paquet, “Planar microlenses for near infrared CMOS image sensors,” in Proceedings IS&T International Symposium on Electronic Imaging 2020, pp. 144-1–144-7(7) (2020).
36. F. Hirigoyen, “Fresnel Zone Plates: plasmonics filtering for SPADs sensors,” in Proceedings of IISW, P45, 222–225 (2017). https://www.imagesensors.org/Past%20Workshops/2017%20Workshop/2017%20Papers/P45.pdf.
37. S. Enoch, A. Gola, P. Lecoq, and A. Rivetti, “Design considerations for a new generation of SiPMs with unprecedented timing resolution,” J. Instrum. 16(2), P02019 (2021). [CrossRef]
38. S. Uenoyama and R. Ota, “40×40 Metalens Array for Improved Silicon Photomultiplier Performance,” ACS Photonics 8(6), 1548–1555 (2021). [CrossRef]
39. L. Dilhan, J. Vaillant, Q. Abadie, A. Ostrovsky, S. Verdet, P. Calka, F. Hemeret, and L. Masarotto, “Planar microlenses applied to SPAD pixels,” in Proceedings of IISW, P14 (2021). https://imagesensors.org/Past%20Workshops/2021%20Workshop/2021%20Papers/P14.pdf.
40. E. Mikheeva, J.-B. Claude, M. Salomoni, J. Wenger, J. Lumeau, R. Abdeddaim, A. Ficorella, A. Gola, G. Paternoster, M. Paganoni, E. Auffray, P. Lecoq, and S. Enoch, “CMOS-compatible all-dielectric metalens for improving pixel photodetector arrays,” APL Photonics 5(11), 116105 (2020). [CrossRef]
41. S. A. Lindner, C. Zhang, I. M. Antolovic, J. Mata Pavia, M. Wolf, and E. Charbon, “Column-Parallel Dynamic TDC Reallocation in SPAD Sensor Module Fabricated in 180 nm CMOS for Near Infrared Optical Tomography,” in Proceedings of IISW, P06, 86–89 (2017). https://www.imagesensors.org/Past%20Workshops/2017%20Workshop/2017%20Papers/P06.pdf.
42. C. Zhang, S. A. Lindner, I. M. Antolovic, M. Wolf, and E. Charbon, “A CMOS SPAD Imager with Collision Detection and 128 Dynamically Reallocating TDCs for Single-Photon Counting and 3D Time-of-Flight Imaging,” Sensors 18(11), 4016 (2018). [CrossRef]
43. S. A. Lindner, C. Zhang, I. M. Antolovic, M. Wolf, and E. Charbon, “A 252 × 144 SPAD Pixel Flash Lidar with 1728 Dual-Clock 48.8 ps TDCs, Integrated Histogramming and 14.9-to-1 Compression in 180 nm CMOS Technology,” in Proceedings of IEEE Symposium on VLSI Circuits (2018), pp. 69–70.
44. C. Zhang, S. A. Lindner, I. M. Antolovic, J. Mata Pavia, M. Wolf, and E. Charbon, “A 30-frames/s, 252 × 144 SPAD Flash LiDAR With 1728 Dual-Clock 48.8-ps TDCs, and Pixel-Wise Integrated Histogramming,” IEEE J. Solid-State Circuits 54(4), 1137–1151 (2019). [CrossRef]
45. A. C. Ulku, C. Bruschini, I. M. Antolovic, Y. Kuo, R. Ankri, S. Weiss, X. Michalet, and E. Charbon, “A 512 × 512 SPAD Image Sensor With Integrated Gating for Widefield FLIM,” IEEE J. Sel. Top. Quantum Electron. 25(1), 1–12 (2019). [CrossRef]
46. A. C. Ulku, A. Ardelean, I. M. Antolovic, S. Weiss, E. Charbon, C. Bruschini, and X. Michalet, “Wide-field time-gated SPAD imager for phasor-based FLIM applications,” Methods Appl. Fluoresc. 8(2), 024002 (2020). [CrossRef]
47. G. Lubin, G. Yaniv, M. Kazes, A. C. Ulku, I. M. Antolovic, S. Burri, C. Bruschini, E. Charbon, V. J. Yallapragada, and D. Oron, “Resolving the Controversy in Biexciton Binding Energy of Cesium Lead Halide Perovskite Nanocrystals through Heralded Single-Particle Spectroscopy,” ACS Nano 15(12), 19581–19587 (2021). [CrossRef]
48. G. Lubin, R. Tenne, A. C. Ulku, I. M. Antolovic, S. Burri, S. Karg, V. J. Yallapragada, C. Bruschini, E. Charbon, and D. Oron, “Heralded Spectroscopy Reveals Exciton−Exciton Correlations in Single Colloidal Quantum Dots,” Nano Lett. 21(16), 6756–6763 (2021). [CrossRef]
49. S. Burri, H. Homulle, C. Bruschini, and E. Charbon, “LinoSPAD: a time-resolved 256 × 1 CMOS SPAD line sensor system featuring 64 FPGA-based TDC channels running at up to 8.5 giga-events per second,” Proc. SPIE 9899, 98990D (2016). [CrossRef]
50. S. Burri, C. Bruschini, and E. Charbon, “LinoSPAD: a compact linear SPAD camera system with 64 FPGA-based TDC modules for versatile 50 ps resolution time-resolved imaging,” Instruments 1(1), 6 (2017). [CrossRef]
51. M. Kollep, “Microlens simulation and fabrication for SPAD arrays,” EPFL-AQUA Semester project (2020).
52. G. Basset, A. Luu-Dinh, C. Schneider, D. Schlup, S. Perrin, W. Fangting, and S. Lecler, “Fabrication of arrays of quasi-micro-beads for massively parallel nanojet super-resolution imaging,” Proc. SPIE 11368, 1136803 (2020). [CrossRef]
53. C. Veerappan and E. Charbon, “A Low Dark Count p-i-n Diode Based SPAD in CMOS Technology,” IEEE Trans. Electron Devices 63(1), 65–71 (2016). [CrossRef]
54. F. Gramuglia, M.-L. Wu, C. Bruschini, M.-J. Lee, and E. Charbon, “A Low-noise CMOS SPAD Pixel with 12.1 ps SPTR and 3 ns Dead Time,” IEEE J. Sel. Top. Quantum Electron. 28(2), 1–9 (2021). [CrossRef]
55. G. Zappalà, F. Acerbi, A. Ferri, A. Gola, G. Paternoster, N. Zorzi, and C. Piemonte, “Set-up and methods for SiPM photo-detection efficiency measurements,” J. Instrum. 11(8), P08014 (2016). [CrossRef]
56. S. K. Yang, J. Lee, S.-W. Kim, H.-Y. Lee, J.-A. Jeon, I. H. Park, J.-R. Yoon, and Y.-S. Baek, “Precision measurement of the photon detection efficiency of silicon photomultipliers using two integrating spheres,” Opt. Express 22(1), 716–726 (2014). [CrossRef]
57. OrmoComp® transmittance curve, micro resist technology, 2022. https://www.microresist.de/en/produkt/ormocomp/. Last visited in Jan 2023.
