Vertical integration of microchips by magnetic assembly and edge wire bonding

The out-of-plane integration of microfabricated planar microchips into functional three-dimensional (3D) devices is a challenge in various emerging MEMS applications such as advanced biosensors and flow sensors. However, no conventional approach currently provides a versatile solution to vertically assemble sensitive or fragile microchips into a separate receiving substrate and to create electrical connections. In this study, we present a method to realize vertical magnetic-field-assisted assembly of discrete silicon microchips into a target receiving substrate and subsequent electrical contacting of the microchips by edge wire bonding, to create interconnections between the receiving substrate and the vertically oriented microchips. Vertical assembly is achieved by combining carefully designed microchip geometries for shape matching and striped patterns of the ferromagnetic material (nickel) on the backside of the microchips, enabling controlled vertical lifting directionality independently of the microchip’s aspect ratio. To form electrical connections between the receiving substrate and a vertically assembled microchip, featuring standard metallic contact electrodes only on its frontside, an edge wire bonding process was developed to realize ball bonds on the top sidewall of the vertically placed microchip. The top sidewall features silicon trenches in correspondence to the frontside electrodes, which induce deformation of the free air balls and result in both mechanical ball bond fixation and around-the-edge metallic connections. The edge wire bonds are realized at room temperature and show minimal contact resistance (<0.2 Ω) and excellent mechanical robustness (>168 mN in pull tests). In our approach, the microchips and the receiving substrate are independently manufactured using standard silicon micromachining processes and materials, with a subsequent heterogeneous integration of the components. Thus, this integration technology potentially enables emerging MEMS applications that require 3D out-of-plane assembly of microchips.


Microchip designs and size variations
After preliminary testing of several trench geometries for edge wire bonding, two silicon trench designs were characterized and proved reliable, both in terms of contact resistance and mechanical fixation. In particular, while several combinations of shapes and dimensions were evaluated, two trench geometries provided the simplest and most reliable solutions. The first geometry is based on a single trapezoid trench, as illustrated in Fig. S1a (design A, presented in the main article). The second design, with a double-trapezoid trench geometry (Fig. S1b, design B), aimed at increasing the potential metallic contact area between the deformed Au free-airballs (FABs) and the frontside electrodes. The important dimensions of the two designs are listed in Table S1, and are defined according to the illustrations in Fig. S1. Fig. S1c illustrates the geometry of the sensing microprobe demonstrator, a possible targeted application that requires the microprobe to be reliably vertically assembled and contacted. The microprobe demonstrator featured top trench geometries similar to design B, but with a top sidewall trench opening width of 50 µm only. The bonded FABs were accordingly scaled down to 70 µm in diameter. The thickness of all presented microchips was 70 ± 5 µm.  Table S1. Design A has a single trapezoid geometry, whilst design B has a double trapezoid shape for enhanced metallic contact area between the deformed Au FABs and the frontside electrodes. (c) Geometrical details of the microprobe demonstrator, fitting inside the lumen of a hollow silicon microneedle. The top part of the "T" shape is narrower, close to the miniaturization limit to perform reliable wire bonding.

Nickel patterning and microchip size variations
Microchips with different aspect ratios (microchip length/width, with the microchip body length fixed to 750 µm) were tested to verify that the lifting direction can be controlled independently of the microchip size or aspect ratio, after Ni patterning. Fig. S2a-d show the backside of the four different microchips, with different widths, after Ni striping. Fig. S2e shows the silicon tabs physically connecting each microchip with the silicon wafer, left during the dry etching step defining the microchips' perimeters, in order to preserve mechanical stability during the following microfabrication steps.

Probability distribution of magnetically assembling individual holes in an array
The magnetic assembly of microchips into the receiving holes was statistically investigated for five different array sizes (one, two, three, four, and six-hole arrays). For each studied array size, the assembly test was repeated 50 times using 39 500-µm-wide microchips. Each progressive step of filling an array had an exponential distribution. A typical example is shown in Fig. S3, where consecutive steps of filling the 6-hole array are presented with their fitted exponential distributions. Figure S3.

Completely filling arrays of different sizes
The probability-density distributions of completely filling arrays of different sizes can be modeled as hypoexponential distributions (Fig. S4). The hypoexponential distributions were created using the exponential distributions fitted to each consecutive step of filling an array of holes. As expected, the mode of a hypoexponential distribution shifts right with an increasing array size (Fig. S4f). Figure S4.

Effects of gap size, microchip number and rotational freedom on assembly efficiency
In order to study the effects of rotational freedom, we performed assembly tests using different microchip widths and different widths of gaps between a microchip and the side walls of a receiving hole. Fig. S5 shows the effects of different chip widths, gap sizes (tolerance between the microchip and the matching receiving hole) and number of microchips used for the assembly test on the resulting assembly efficiency, in accordance with the different degrees of rotational freedom. Fig. S5f shows that larger gap sizes and narrower chips tend to increase the assembly efficiency. A larger number of microchips can increase the assembly efficiency as well. However, this latter effect is less evident than the previous two, probably because, regardless of the number of microchips available, only a limited number of them can move over a single hole in a single magnet sweep. This effect would likely be evident when the total number of chips approaches or equals the number of holes. The effects on the assembly efficiency of both gap size and microchip width can be explained by looking at the range of rotational angles of the microchips that allows them to drop into the receiving holes (Fig. S5g). By using the rotational freedom, assembly speeds for different microchip and receiving hole size variations can be forecasted. By using the available rotational angles as a scaling factor for the mean number of magnet sweeps, forecasts for the probability density distributions of 500-µm-wide microchips with 10 µm gap and 330-µm-wide microchips with 20 µm gap have been made (Fig. S5f), based on the assembly data from 500-µm-wide microchips with 25 µm gap. These forecasts are in accordance with the measured data (Fig. S5f). This indicates that the range of allowable rotational angles of the microchips to fit in the receiving holes has a major effect on the assembly efficiency. A circular receiving hole would therefore significantly increase the assembly efficiency. Faster assembly into circular holes might be preferable in certain applications and it would be compatible with the electrical contacting technique presented hereafter. However, for applications where the rotational positions of assembled microchips are important, rectangular holes serve a role for aligning the microchips and, additionally, can save area on the receiving substrate.  Table S2 reports the parameters to perform wire bonding on microchips with design A and B. The minimum force needed for placing a ball bond is larger for design B, since more deformation of the FAB into the trenches has to be induced for reliable mechanical fixation. Fig. S6 shows the edge wire bonding results for design B from different viewing angles.

Pull tests
The measured mean pulling strength for breaking the wire bonds in any location was 92.1 mN, with a minimum measured pulling force for breaking a wire bond of 63.7 mN. In fact, all the broken points occurred at the stitch bonds on the receiving substrate and not on the ball bonds on the microchip side. These values were nonetheless almost four times larger than 23.5 mN required in military standard, showing more than sufficient mechanical strength of the wire bonds. Even after stich bond fixation by epoxy glue, all the tested bonded Au wires were torn before occurrence of any ball-to-trench bond failure.

Electrical characterization
The contact resistance was also studied for both designs, by comparing the measured resistance of the thinfilm resistors before and after bonding. The average increase in the resistance values for design A and B are 0.19 Ω and 0.08 Ω, respectively, which include the contribution from the contact resistances of the edge ball bonds. The average increase in resistance is lower for design B than design A, even though both values are significantly smaller than the standard errors of the sample mean. Therefore, the contact resistances of the edge ball bonds for both microchip designs are negligible. The main advantage of design B over design A is related to the increased Au-to-Au bond area, which could be seen in the SEM inspection (c.f. Fig. 4c versus Fig.  S6c), potentially resulting in improved reliability in a production environment. Table S3 summarizes the results of the resistance measurements.