1. Introduction

For any electromagnetic (EM) wave sensor to work, it must be illuminated by the signal to be detected. This prerequisite poses a challenge when one uses a micro or nano sensor to detect optical signals in an optic fiber, or any EM signals in a micro waveguide, since it is difficult to accurately align two micron-sized or smaller objects, especially in applications where bulky automatic alignment equipment cannot be employed and in situ adjustment after the alignment is not possible. Meeting this challenge is particularly important for single-photon or photon-number-resolving detectors such as superconducting transition-edge sensors (TESes) [1], as any misalignment leads to photon loss that lowers the detection efficiency (DE) critical for these detectors.

Due to the importance of alignment, it has been an area of active research in the forefront of optical senors and optoelectronics. For fiber-coupled single-photon or photon-number-resolving detectors, poor input signal coupling can result due to partial overlap between the sensor’s active area and the beam spot of the fiber, as well as an excessive gap between the sensor and the fiber tip that leads to light beam divergence. In the context of TES, it is possible to position an optical fiber using a V-groove and align it with a small TES by visual inspection under a microscope [2]. However, a beam spot size larger than the TES caused by a relatively large gap between the fiber tip and the TES, as well as any error in visual alignment, results in a poor DE. In order to improve the accuracy of alignment, one can use an inverted infrared microscope to observe the image of the infrared light beam coming out of the optical fiber from the backside of the chip and adjust the position of the fiber until the beam spot coincides with the TES [3]. Such a technique provides definitive evidence for alignment, but it requires the use of specialized imaging equipment that enables observation at wavelengths the chip substrate is transparent to. In another technique [4] the authors called “self-aligning," a circular area around the TES about the size of an optical fiber is etched off the wafer and fit into a commercially available optical fiber ferrule sleeve where it is in alignment with an optical fiber inserted in the ferrule. Though accurate and reliable, this technique employs a time-consuming and costly process to etch through the entire depth of a wafer, which offsets the benefit and appeal of exploiting a commercial optical fiber ferrule assembly. Also, since the TES is etched into a discrete device, it cannot be used for alignment of on-chip sensor arrays, a disadvantage for integration considerations. As another notable challenge, it is difficult in many alignment schemes to precisely control the gap between the fiber tip and the sensor which is a critical parameter [5]. A large gap leads to a beam spot larger than the sensor size and results in coupling loss. On the contrary, if the fiber tip is positioned too closely to the sensor, one risks damaging the sensor during the alignment process, or when the sensor is cooled down and components in the alignment assembly contract at lower temperatures.

Considering the drawbacks of presently available methods, we are motivated to explore new techniques that can accomplish accurate alignment for highly efficient fiber-sensor coupling, yet are much less demanding in equipment, fabrication, implementation, and cost. We are also interested in finding easy and reliable verification methods to evaluate the performance of our alignment techniques, without having to resort to advanced equipment and dedicated experimental setups for through-chip observation at infrared wavelengths.

2. Alignment via micro-structures on the surface of the sensor chip

The underlying principle of our alignment technique is illustrated in Fig. 1. In essence, we fabricate a circular alignment structure, designated as the "aligner," on the surface of the sensor chip with the sensor at its center. The inner diameter of the aligner is slightly larger than that of an optical fiber, such that a fiber can be inserted in it. It is easy to see that, once the optical fiber is inserted in the circular structure and aligned with it, the fiber core at its center is also aligned with the sensor located at the center of the aligner. This basic idea allows us to align the fiber cladding to the circular aligner, instead of aligning the sensor to the fiber core directly. It presents a significant advantage in practice, since observation and manipulation of the fiber cladding, usually as large as 125$\mu$m, is much easier than direct alignment of the fiber core which can have a diameter less than 10$\mu$m for a single-mode optical fiber, though the required alignment precision is comparable. As long as the sidewall of the aligner is vertical, it also allows to push the fiber tip very close to the sensor without misalignment, once the fiber cladding is inserted in the aligner. This is a major advantage because, if one tries to align the fiber core to the sensor directly, both the sensor and the beam spot of the optical fiber will be blocked from the microscope view by the fiber cladding when the fiber tip is very close to the chip. Since the aligner can be easily replicated for all sensors on a chip in a single fabrication process, our technique supports alignment of on-chip sensor arrays with densely integrated devices in a compact package.

 

Fig. 1. Alignment structure (aligner) and proximity structure (spacer) on the surface of the sensor chip. (a) Top view without the optical fiber. (b) Sectional view with the optical fiber along the dashed line in (a).

Download Full Size | PDF

In Fig. 1, there are also proximity structures which we call "spacers" on the chip surface that will prevent the fiber tip from contact with the sensor. The spacers should be fabricated within the inner perimeter of the circular aligner, but with a distance from the center greater than the radius of the fiber core, such that they will come into contact with the fiber cladding to stop the optical fiber when it moves toward the sensor, without blocking light coming out of the fiber core. The proximity structures can be fabricated together with other parts of the sensor such as the lead wires. In fact, it is possible to use the lead wires as spacers as long as their thickness and locations are suitable. With these simple proximity structures, the gap between the fiber tip and the sensor can then be set precisely at their height if one pushes the optical fiber to the end of the alignment structure. By choosing a small height for the spacers, the fiber tip can be safely positioned very close to the sensor, allowing for efficient coupling of the optical signal, without using any specialized equipment and advanced technology to measure and position the fiber tip at such a small gap.

Obviously, the principles of our alignment scheme can be realized by many micro fabrication and surface micro machining techniques, as long as the process employed can reach a precision required for the alignment of the sensor and fiber core with given sizes. Since the resolution of modern optical lithography and micro fabrication can reach the submicron level, such a requirement is easily met in practice. In order to ease the alignment operation and reduce the uncertainty in positioning the optical fiber, it is preferred to have a deep aligner with nearly vertical sidewalls. To accomplish the alignment, one only need to observe the fiber and the aligner in the microscope from the front side of the chip, and use a 3-axis translation stage to adjust the position of the optical fiber until it is inserted into the aligner. After the alignment, we can glue the optical fiber to the sensor chip to fix its position, by using e.g. a photon sensitive epoxy cured by UV light.

In Fig. 2, it is shown an easy and reliable method to examine the result of alignment without using an infrared microscope. In this method, a dummy sensor used to evaluate the performance of our alignment scheme is fabricated on a substrate transparent to visible light. Also, a chip holder with a through-hole larger than the sensor is made. We then glue the sensor chip to the chip holder, at such a position that the sensor is located within the perimeter of the through-hole, as depicted in Fig. 2(a). After aligning an optical fiber to the sensor as described earlier, we can then examine the result of alignment with the setup shown in Fig. 2(b). Specifically, we place the chip holder on a sample platform with the sensor chip facing sideways, and couple a bright visible laser light into the optical fiber. At this point, we can use an ordinary microscope in the plane of the sample platform to observe the sensor and the beam spot from the backside of the chip via the hole in the chip holder and through the transparent substrate. This allows us to compare their relative positions and verify the result of alignment, without requiring an infrared microscope and supporting equipment for infrared illumination and other necessary optical setups, and thus is much easier to realize and much lower in cost.

 

Fig. 2. (a) Dummy sensor fabricated on a transparent substrate and placed in a through-hole on a chip holder. Alignment and proximity structures omitted for easier viewing. (b) Setup to examine the result of alignment by sideway observation of the sensor and beam spot via the hole in the chip holder and through the transparent substrate.

Download Full Size | PDF

3. Demonstration with dummy sensors

To implement our alignment technique and evaluate its performance in practice, we fabricate dummy thin-film sensors of various sizes on transparent quartz substrates, as shown in Fig. 3. Since these dummy sensors are only used to evaluate the precision of our alignment technique, they can be made in any material and with any process, as long as their sizes and layouts are similar to real devices for which our alignment method is to be applied. In our experiments, they are mostly made together with our TES devices described in section 4 to reduce our workload. Consequently, they share the same material and layout with some minor differences. In brief, the dummy sensors are made by depositing titanium/gold (Ti/Au) thin films on the quartz substrate and subsequently etching the resulting films into small square devices in different sizes between 5 and 20$\mu$m. The niobium lead wires are defined by a lift-off process. They are also the proximity structures to protect the sensor from abrasion caused by the tip of the optic fiber. See section 4 for greater details of the device fabrication process.

 

Fig. 3. (a) Micrograph of a 9$\mu$m-core single-mode optical fiber inserted in the aligner for a dummy sensor fabricated on a quartz substrate. (b) Image of a 5$\mu$m$\times$5$\mu$m dummy sensor, and that of the beam spot when a bright red light is coupled in the optical fiber aligned to the sensor.

Download Full Size | PDF

We then fabricate alignment structures on the surface of the dummy sensor chip. In our demonstration experiment, we simply pattern photoresist into a circular aligner on the surface of the dummy sensor chip which has an inner diameter a couple microns larger than that of an optical fiber. The location of the aligner is so chosen that the dummy sensor is at its center. There are gaps along the perimeter of the aligner to allow lead wires through. In fabricating the aligner, we choose SU-8 [6], a thick photoresist which is widely used in surface micromachining because of its high resolution, excellent mechanical strength and durability, and extraordinary ability to form tall sidewalls that are nearly vertical. In addition, it has the major advantage of being resilient to cryogenic temperatures without cracking [7]. Therefore, it is an ideal choice for cryogenic sensors able to withstand thermal cycles in the refrigerator. As our aligner fabrication involves only cheap but accurate and reliable photoresist patterning, it does not incur any time consuming and expensive process in other solutions such as the self-aligning method [4]. Consequently, it can achieve a comparable or higher resolution at much lower technical complexity and implementation cost.

In Fig. 3(a), it is shown an optical fiber inserted in the alignment structure and aligned to the sensor. The optical fiber is prepared from a fiber optic pigtail with a connector on the input end which allows to couple an optical signal to the sensor easily once it is aligned with the fiber core. The alignment is done by attaching the optical fiber to a 3-axis translation stage which adjusts its position. Following the alignment, we use the method described in section 2 to verify the result of alignment. As shown in Fig. 3(b), devices as small as 5$\mu$m$\times$5$\mu$m are successfully aligned to the 9$\mu$m fiber core.

4. Application to TES

Clearly, our alignment scheme is agnostic to the material and detection mechanism of the sensor. It can then be used for many micro and nano semiconductor and superconductor detectors. As an example of fruitful application, we use this technique to couple optical signals to a TES device designed for multi-photon quantum optics experiments around 830nm. These experiments require a sufficiently high energy resolution to resolve photon numbers at the target wavelength, a high counting rate above 200kHz, and a decent DE for the sensor. To meet these requirements, we choose a Ti/Au double layer [8] structure for the TES, in which the proximity effect at the interface between the layers reduces the critical temperature of the device to around 200mK. At such a temperature, the thermal conductance of the gold layer is sufficiently high to enable a thermal recovery time constant within 1$\mu$s leading to a high counting rate.

The fabrication process of our TES device and its aligner and spacer is illustrated in Fig. 4. We first deposit a 300nm thick layer of silicon nitride (SiN) on a silicon substrate by low pressure chemical vapor deposition (LPCVD). Then, we deposit a very thin titanium layer on the SiN as an adhesion layer, on which a 30nm gold layer is grown, followed by the growth of a 30nm superconducting titanium layer, all using electron beam evaporation in a high vacuum environment of 10$^{-7}$ Torr. In such a Ti/Au double layer structure [8], the proximity effect [1] at the interface between the gold and titanium layers reduces the critical temperature of the device from that of the top superconducting titanium layer. The drop in critical temperature is related to the relative thickness of the normal metal layer and superconducting layer. The layer thickness we choose results in a critical temperature of 200mK. Next, we etch the Ti/Au thin film into small square devices in sizes of 5-20$\mu$m as shown in Fig. 4(d), using reactive ion etching and ion milling. Subsequently, niobium lead wires are grown and patterned together with the proximity spacers by sputter deposition in a lift-off process as depicted in Fig. 4(e). In the standard TES circuit, the niobium leads provide a superconducting connection to the external bias and readout circuitry. Its relatively high critical temperature is advantageous for minimizing heat conduction through the leads. In our design, the lead wires are also part of the proximity structures. Niobium is an excellent material for the spacers to protect the sensor from abrasion as it is very hard and unlikely to deform in contact with the fiber cladding. With a thickness of 250nm, the niobium lead wires and proximity spacers allow us to position the tip of the fiber within 200nm of the TES when the fiber is inserted in and pushed to the end of the aligner.

 

Fig. 4. Fabrication process of the TES device, as well as its aligner and spacer. (a) The silicon substrate. (b) SiN layer deposition by LPCVD. (c) Deposition of the adhesion layer, the gold normal metal layer, and the titanium superconducting layer by ebeam evaporation. (d) Etch of the metal layers into micron-sized TES devices. (e) Fabrication of the niobium lead and spacer in a lift-off process. (f) Lithography and patterning of the SU-8 photoresist into the aligner.

Download Full Size | PDF

The fabrication of the aligner in Fig. 4(f) and the alignment of a 9$\mu$m-core single-mode optical fiber to the TES device is as described in section 3. In Fig. 5(a), a micrograph of a completed device with its aligner is shown. To evaluate the resolution of the aligner fabrication, we measured the diameter of 20 fabricated aligners on SiN and found a standard deviation of 0.62$\mu$m. Such a small error is more than sufficient for the alignment of our square devices that are 5$\mu$m or larger, though the resolution can be further improved by fine tuning the lithography and patterning process for the SU-8 photoresist.

 

Fig. 5. (a) Micrograph of a fabricated TES device, its aligner, and its proximity spacers including the lead wires. (b) The TES biased with a current source $I_b$ and a shunt resistance $Rs$ much smaller than its resistance, and an inductively coupled SQUID measuring current responses to optical signals. Shown in the inset is a typical scope reading of many photon absorption events with a time resolution of 1$\mu$s per division.

Download Full Size | PDF

In choosing TES devices for our optical experiments, we must make compromises between their design goals which are dictated by the intended application. Specifically, we can achieve a higher energy resolution by using a device with a lower critical temperature or a smaller size, however its counting rate will be slower at lower temperatures [1] and its DE will suffer if the sensor is smaller than the beam spot size of the fiber. In our experiments, we must ensure an energy resolution that is sufficiently high to resolve photon numbers at the target wavelength. To evaluate the precision of our alignment method and the coupling efficiency it enables, any photon loss is undesirable, which precludes devices smaller than the size of the fiber core. Therefore, we choose a 10$\mu$m$\times$10$\mu$m device for our optical experiments, with a critical temperature of 200mK and a normal resistance of 0.6$\Omega$. By using a current source and a shunt resistance much smaller than the TES resistance, we bias the device in the voltage mode, as shown in Fig. 5(b). It is operated at a base temperature much lower than its critical temperature to ensure strong eletro-thermal feedback.

In the optical experiments, heavily attenuated laser pulses with a repetition rate of up to 300kHz from an 830nm LED are fed into the TES through a single-mode fiber. As shown in Fig. 5(b), the change in the TES current caused by the absorption of photons is measured by a SQUID inductively coupled to the TES circuit. A typical color temperature display of scope traces produced by many photon absorption events is shown in the inset. With a total optical response time of about 2$\mu$s, the TES can be operated at a counting rate up to 500kHz. In Fig. 6, it is shown the histogram for the amplitude of the response curves collected in an optical experiment which exhibits clear photon number discrimination at the target wavelength. Also, the fit according to a Poissonian emission function is shown. From the standard deviation of the fit, we obtain a FWHM energy resolution of 0.65eV, without any filtering and smoothing of the experimental data. Using the estimated mean photon number from the fit, and the input power to the refrigerator’s optical fiber port measured by a calibrated setup, we obtain a DE of 41%. Since there is no multi-layer optical resonance structure fabricated on our device, we expect an upper limit for the DE determined by the reflectance of the TES material and fiber and connector loss. The measured DE is consistent with the reflectance of surface oxidized titanium [9,10] between 55% and 60% at 830nm, indicating very low coupling loss resulting from accurate alignment between the sensor and the optical fiber. Hence, with the help of our alignment scheme, we achieve all design goals for our TES device.

 

Fig. 6. Amplitude histogram of the photon response curves and its fit by the Poissonian emission distribution function (0-photon peak not shown).

Download Full Size | PDF

5. Conclusion

Our solution to the nontrivial challenge of accurate fiber-sensor alignment is based on the simple idea that alignment is achieved when the fiber is inserted in the aligner, provided that the position of the sensor relative to the aligner coincides with that of the fiber core relative to the cladding. In comparison with conventional methods, it has many advantages. Not only does it achieve a high alignment precision enabled by lithography, but it is very easy to implement as it only involves simple and reliable photoresist patterning and does not require any specialized imaging equipment and intricate fabrication process. It supports alignment of densely integrated on-chip detectors due to the replicable nature of the micro-fabricated aligners. Therefore, our method can dramatically lower the technical barrier and implementation cost for fiber-sensor alignment to help fully realize the potential of optical microsensors. Since our technique does not depend on the material and detection mechanism of the sensor, it can be adopted in many sensor technologies [11,12] to overcome the difficulty of efficient input coupling, in its original form or with necessary modification pertinent to the specific application. It is even possible to use it for efficient light collection in light emitting devices [13,14].