Fatigue Cycling of Electrical Interconnects Dispensed on Flexible Substrate

Recently, flexible electronics have attracted significant attention in many applications such as smart phones, human performance wearable monitoring electronics, Internet of things, solar cells, and sensors that can be used for various applications including healthcare, aerospace, and automotive [1]. Flexible electronics are circuits fabricated on thin, compliant substrates such as polyimide (PI), polyethylene terephthalate (PET), thermoplastic polyurethane (TPU), and paper [1]. The advantage of flexible electronics over conventional electronics is that they can be twisted, bent, folded, and stretched without significant effect on their functionality [1]. In addition, flexible electronics are light weight, have low fabrication costs, and compatible with rapid fabrications and high throughput (roll to roll fabrication) operations [1]. Conventional electronics components, such as batteries, integrated circuits, interconnects, resistors, capacitors, and transistors, etc, can be integrated on to a flexible substrate, resulting in flexible hybrid electronics (FHE). Flexible electronics can be fabricated either by conventional microfabrication techniques (e.g. lithography) or printing techniques (e.g. inkjet printing). However, fabrication of flexible electronics using conventional techniques has limitations in terms of the cost associated with it and the complex multi-step processes involved. Furthermore, many flexible substrates are not compatible with microfabrication processes. The advantages of printing technologies over conventional techniques include high throughput, low cost, rapid prototyping, low temperature process, and wide range of materials that can be used [1, 2]. Dispenser printing, which is used in this work, is an additive manufacturing process that directly deposits dots or lines of material based on a programmed tool-path. This technique is capable of printing complex structures using a wide range of functionalized inks, like those used in screen printing while removing the need to invest in a screen. Dispenser printing offers design flexibility and less ink waste, as ink is dispensed where desired compared to screen printing. These advantages make the dispenser printing attractive, especially where the design often changes or rapid prototyping is required.

Electrical interconnects are an important part of any electronics circuit including FHE circuits to enable the connection between different circuit components. Failure of these interconnects lead to product failure, so understanding the reliability of them is very important. In addition, the material selection is a very crucial and challenging step in design and fabrication of printed electronics circuits. Therefore, we present this work as a systematic guideline that will be helpful for both flexible electronics manufacturers and researchers in understanding the behavior of printed interconnects in response to fatigue cycling. To achieve this objective, two materials were dispensed as fine conductive interconnects (traces) on a flexible substrate. The damage in the printed traces caused by mechanical tension was assessed by in-situ monitoring of the change in the resistance of traces during fatigue cycling. In addition, a novel method for test results interpretations and understanding is introduced.

As is shown in Fig. (1A) a 3-axis robot dispenser system, Nordson EFD PRO4L/B, connected to a pressure, time, and vacuum controller was used for printing. Material was loaded into a syringe which is mounted on the robot head. The system uses compressed air to create pressure to the syringe and expel material through a needle. The robot head moves in x-z directions while the substrate is placed on a movable (y direction) vacuum platen. The robot moves based on programmed paths controlled by motion software. CCD camera was added to the system to accurately set the tip to the substrate distance. The features of the printed structures are controlled by the needle diameter, printing speed, pressure, vacuum, and distance between substrate and the needle tip (stand off distance) at room temperature.

Two thick pastes, SW 1400 from Asahi Co. and 125-13 HV from Creative Materials Co., were utilized to dispense fine electrical traces. The test vehicle shown in Fig. (2A) was dispensed on a 50 μm thick UBE UPILEX-50S polyimide substrate. The test vehicle was designed to have four 1×1 mm contact pads to enable four wires resistance measurements and 31 mm sense line.

After dispensing, SW 1400 silver paste traces were cured at 150°C for 20 minutes in a convection oven while the 125-13 HV traces were cured at 170°C for the same time, following manufacturers recommendation. Table (I) summarizes manufacturer specifications of the materials.

The dispensed traces were subjected to fatigue cycling using Instron 3344 Single Column Load frame with 2 kN load capacity where the trace was pressure-fixed between two mechanical grips attached to the load column as is shown in Fig. (1B). The electrical resistance of the traces was measured in-situ using a Keithley 2100 series multi meter during mechanical testing. To eliminate the effect of the contact pads and the meter wires resistances, four wire resistance measurements were performed by attaching four copper leads to the trace pads using conductive adhesive epoxy from Chemtronics. The fatigue cycling tests were performed at 2.5%, 2.0%, and 1.5% strain amplitude for 300, 600, and 1000 cycles, respectively, at 0.05 mm/s displacement rate at room temperature. These strain values have been chosen to cover the possible use case conditions and accelerated testing.

First, the dispenser parameters were optimized to achieve fine line features through statistical design of experiments approaches. The optimization process included screening experiments, factorial design, and surface response optimization. The finest line had a width of 160–175 μm and thickness of 9–12 μm obtained using the optimal parameters listed in Table (II) obtained via the optimization studies.

Table (III) shows the resistance statistics of the dispensed traces. The sheet resistance was calculated from the measured four wires resistance, width, and thickness. As expected from the data sheets, SW 1400 traces had higher resistance due to lower silver percentage loading. The energy-dispersive X-ray spectroscopy (EDS) measurements, Table (IV), confirmed as well that the silver percentage of SW 1400 trace was lower than 125-13 HV. The agreement between the calculated sheet resistance with the suppliers specs and the small variations between the dispensed traces support the feasibility of using dispenser system for printed flexible electronics applications.

By comparing the dispensed structures and the cross sections of both materials, as is shown in Fig. (3), the 125-13 HV traces showed better line quality with minimal rough edges and uniform cross section.

Before discussing the behavior of the dispensed traces on flexible substrate under fatigue cycling, it is important to point out the role of the flexible substrate. The substrate supports the printed structures and carries most of the deformation during twisting, bending, or stretching. In other words, here in fatigue cycling tests, the load is directly applied to the substrate (flex) not to the dispensed trace.

Fig. (4a) shows an increase in the resistance of the trace while stretching to 2.5% of the trace length between the grips. This change in the resistance is due to the effect of strain, cracks induced, and increase in the inter particle distances as a response to the stretching. However, inspecting traces after testing showed no remarkable change or cracks. As is shown, the change in the resistance of the SW 1400 trace was higher than that of the 125-13 HV trace. Furthermore, the change in the resistance of 125-13 HV trace is very small until 0.5% strain and then starts increasing rapidly. Overall, the change in the trace resistance is about 10% at the peak strain (2.5%) for both materials. Fatigue cycling of dispensed traces shows an increase in the resistance from one cycle to another as shown in Fig. (4b).

To understand the change in the electrical resistance with the cycle counts during cycling, behavior of trace in the first cycle is explained in some of details as follow. Let A₀, L₀, and R₀ represent cross sectional area, length, and resistance of unstrained trace, respectively, whereas A, L, and R have same representation of strained trace. During stretching, there is no change in the resistivity of the material or in the total volume of the trace based on assumption that has proved using experimental and simulation studies [3]. Based on that, the total volume of unstrained trace should be equal to the volume of the strained trace which is represented mathematically by A₀L₀ = AL. Substitution of A from R = ρ * L/A into the last expression, where ρ is the material resistivity, yields (1).

Fig. (5) represents the relative resistance (RR) of strained trace to 2.5% versus square of relative length in the first half of the first cycle (during loading). It is noticed that the experimental fitted curve of (1) deviates from the theoretical curve. This behavior was noticed from the first cycle for both materials. This deviation is explained by crack induced during stretching from the first cycle [3]. However, during unloading [Fig. 6a] there is decrease in the resistance, which is explained by closing the cracks induced while loading; particles come closer, but do not heal. Moreover, the final trace resistance during unloading is not the same as the initial resistance during the loading part which explains damage occurred to the trace.

Fig. (6b) shows an example load vs. strain curve, with the sample showing zero load before all strain is recovered. This phenomena is explained by the viscoelastic behavior of the polyimide. The viscoelasticity is a material (polymer) property which means that the dispensed traces have no or negligible effect on the viscoelastic strain as confirmed in Fig. (7a) for both materials used in this study. But indeed, the viscoelasticity is strain amplitude dependent, and as the strain amplitude increases as the viscoelastic strain is increased [Fig. (7b) for SW 1400 trace at 2.5%, 2.0%, 1.5%, and 1.0% for 300 cycles]. As shown Fig. (7a) and Fig. (7b), the viscoelastic strain increases with the number of cycles and then remains almost constant.

Fig. (8) shows accumulation of change in the trace resistance with cycle count at strain of 2.5% each cycle where the change in the resistance of SW 1400 trace is higher. Due to viscoelastic behavior of polyimide, the initial resistance has been measured at 0.5% strain [4], which is higher than the viscoelastic strain of polyimide for both dispensed materials at the last cycle.

Fig. (9a) shows the electrical resistance of SW 1400 traces at 0.5% strain while loading at each cycle for 300 cycles. It is clear that the higher strain induces more damage, meaning larger change in the resistance. Regardless of the strain amplitude, the shapes of the curves in Fig. (9a) look similar with different slopes, meaning that any curve could be scaled into the other with some factor. Fig. (9b) shows scaled curve at 1.5% to 2.5% strain amplitude with scaling factor of 5.386, signifying that the damage occurred at 2.5% strain is 5.386X faster than damage occurred at 1.5% strain amplitude. Following the same approach, damage at 2.0% is 3.266X faster than at 1.5% strain amplitude and damage at 2.5% is 1.49X faster than at 2.0% strain amplitude.

Following same approach for 125-13 HV traces, Fig. (10a) shows the same conclusion as SW 1400 traces; higher strain causes stronger change in the resistance at 0.5% strain during loading for 300 cycles. Damage at 2.5% is 6.289X faster than at 1.5% strain amplitude [Fig. (10b)], at 2.0% is 3.727X faster than at 1.5%, and at 2.5% is 1.421X faster than at 2.0% strain amplitude.

Fig. (11a) compares the change in the resistance of both materials at 0.5% strain while loading when the traces cycled at 2.5%, 2.0%, and 1.5% strain amplitude for 300 cycles. The change in the resistance of SW 1400 traces is stronger than that of 125-13 HV traces. The damage in SW 1400 traces is 1.111X, 1.218X, and 1.624X faster than 125-13 HV traces at 2.5%, 2.0%, and 1.5% strain amplitude, respectively. Fig. (11b) shows scaled resistance change with number of cycles of both materials at 2.0% strain amplitude. Although, the difference is not large, it is significant for applications such as human vital signs monitoring devices.

The Scanning Electron Microscope (SEM) images in Fig. (12) depict the surface and the cross sectional morphology of traces after curing. SW 1400 traces have bigger flakes and larger particles while the 125-13 HV traces have smaller and finer particles. Moreover, the cross sectional morphology shows that the 125-13 HV traces have higher micro structure density, more particle aggregation, and significantly less inter particle distance compared to SW 1400 traces. During the fatigue cycling, the mechanical load increases the inter particle distance, damaging the entanglement structure of the polymer network and causing particles to move away from each other (dislocation and crack formation) which caused increased resistance during fatigue cycling, [5]. However, inspection tested samples microstructure showed no evidence of testing, meaning no clear cracks which could be due to relaxation of the flexible substrate. Micro structure differences make 125-13 HV traces more reliable than SW 1400 traces.

Coffin-Manson model is one of the best known and widely used models for isothermal fatigue cycling that relates the life time with the plastic strain in cycling; given by (2).

where Δε_p is the plastic strain amplitude, ε′_p is the fatigue ductility coefficient, N_f is the number of cycles to failure, and c is the fatigue ductility exponent. In general, Coffin-Manson model is applicable in the low cycle fatigue regime at plastic strain. Polymers are a complex case, since the total strain is composed of plastic, elastic, and viscoelastic strain. Based on that, the total strain is used in this study to model the trace life time since measuring the plastic strain during the tests was not possible.

Considering 20% increase in the initial resistance while fatigue cycling as a failure criteria, the fatigue life time of the dispensed samples was modeled using the Coffin-Manson isothermal fatigue model. Equation (2) was fitted by using the number of cycles to failure (20% increase in the resistance) and the total strain [Fig. (13)]. The goodness of fitness was assessed by calculating the R-squared and the root mean square (RMS) for the fitted model with 95% confidence interval. R-squared explains the variation in the data and it takes values between 0 and 1; where 1 means greater proportion of variance is accounted for by the model. Root mean squared measures how close the actual data to the predicted data by the model; where the closer to zero, the better fitted line. R-squared was found to equal 0.932 and 0.957 for 125-13 HV and SW 1400. Moreover, RMS was found to equal 0.041 and 0.033 for 125-13 HV and SW 1400. The constants of the Coffin-Manson equation were determined empirically from fitted line of (2). As expected from (2), Fig. (13) shows decrease in the life time as strain increases. The fatigue ductility exponent (c) was found to be −0.2531 and −0.2968 for 125-13 HV and SW 1400 dispensed traces, respectively. The fatigue ductility exponent was reported in the literature to be in the range of −0.5 to −0.7 for the bulk metals [6]. As a result of using the total strain instead of the plastic part, smaller fatigue ductility exponents were obtained in this experiment. Moreover, smaller values were reported for thin films; sputtered copper and silver films [5], Roll-to-Roll printed silver pastes [7], and sputtered copper and aluminum films [8]. Based on the results reported in this section, Coffin-Manson model could be applicable for printed traces but more investigation on what the percentage of the used total strain can be attributed to the plastic strain is required.

Fatigue cycling of dispensed traces caused damage from the first cycle and accumulated with cycle count. Micro structure of the materials affected the behavior of the trace under fatigue cycling. The material with finer structure and shorter inter particles distances is more durable during fatigue cycling. In printed flexible electronics, generalization of reliability results is challenging due to role of flexible polymer substrate. These results could be different when a different substrate or different fabrication technique is used. Moreover, the behavior of printed traces is dependent on many factors such as strain rate, mechanical and thermal loading history etc. The life time of printed interconnects was successfully modeled using the Coffin-Mansion model using the total strain. This approach could be applicable to a some extent however other models could be more representative.

This material is based, in part, on research sponsored by Air Force Research Laboratory under agreement number FA8650-15-2-5401 via FlexTech Alliance, Inc., as conducted through its flexible hybrid electronics manufacturing innovation institute. The U.S. government is authorized to reproduce and distribute reprints for governmental purposes notwithstanding any copyright notation thereon.