A highly miniaturized wireless inertial sensor using a novel 3D flexible circuit

A. System overview

Figure 2 shows the most basic method in remotely capturing motion data and transmitting it to a receiver or base station.

Inertial sensors (e.g. accelerometers, gyroscopes and magnetometers) sense movement and the sensor outputs (assuming analog outputs) are applied to a signal conditioning block. The signal conditioning stage usually has some form of amplification and filtering that are required to increase the amplitude and band‐limit the sensor output signals prior to analog‐to‐digital (A/D) conversion. In modern systems, an A/D conversion is carried out as soon as possible, to take advantage of the many possibilities that are offered by the ever‐increasing power of digital electronics.

The transmitter stage sends the acquired and processed data to a remote receiver. At the receiver side, the signal is decoded, processed and the information of interest is made available to the user in a graphical user interface (GUI) for example. In order to implement the wireless inertial sensing system, a number of functional blocks are required as shown in Figure 3.

At the inception of this project, only single‐axis gyroscopes were commercially available and therefore three gyros are required to provide the required three DOF. The analog output voltages need to be digitized and therefore a convenient and space saving method is to use a microcontroller (MCU) with on‐board analog‐to‐digital converter (ADC). A programming connector is required to allow the user program to be downloaded to the MCU. A radio integrated circuit (IC) and antenna are also required to wirelessly transfer the captured motion data to another receiver node or base station. A battery and power subsystem is required to provide system power.

Apart from the obvious small size requirement of the WIMU cubed system, a very important aspect of the design is to ensure low‐power operation. To facilitate this requirement, each sub‐system element needs to use as little power as possible during operation as well as in standby/sleep mode. MEMS gyroscopes in particular are power hungry devices and having the ability to enable or disable the power to the gyroscopes is something that is desirable.

B. 3D package implementation

In order to realize a 3D cubed system, a number of different packaging strategies were considered. In particular, the use of a flex material was chosen as the focus of this work. The chosen concept is to fold a six‐face crucifix shaped FPC into a cube as shown in Figure 4. A functional system block is assigned to each face, with the flexible PCB providing the means of interconnecting each of these faces.

For this system, each face of the crucifix is 10 mm² and the total size is 40×30 mm in length. The outer and inner layers of each are then assigned a specific function as shown in Figure 5.

The assignment of functions to faces was chosen as follows: single‐axis gyros are mounted on three adjacent faces as shown in Figure 5(a) such that when folded, they are mutually orthogonal. A tri‐axial accelerometer is mounted on an adjacent face. The radio transceiver IC and antenna are placed on adjacent faces to minimize radio frequency (RF) losses between transceiver and antenna. In addition, when the device is folded, the antenna is placed on the outside surface of the cube to maximize the efficiency of the antenna. The surface behind the antenna is deliberately kept free of components so as not to adversely affect the antenna impedance and radiation characteristics. The MCU is assigned to another face and in order to conveniently program the device, a programming connector is assigned to the same face but on the outside layer to allow the MCU to be programmed without having to open the folded cube. The majority of discrete passive components such as capacitors, inductors and resistors are placed in the centre face, close to the four adjacent faces, easing the routability of the interconnect.

C. Sensor system components

Having defined the functional blocks for the system and the manner in which these blocks have been partitioned on the substrate, the choice of components used to implement the system are now described in detail.

The chosen gyroscope for this application system was the ADXRS150±150°/s MEMS gyroscope with on‐chip signal conditioning (Analog Devices, n.d). The device has a package size of 7×7×3 mm and allows the gyroscopes to fit comfortably in the centre of each of the three 10 mm² cube faces on the flexible substrate. The gyro output signal, RATEOUT, is a voltage proportional to the angular rate about the axis normal to the top surface of the package as shown in Figure 6.

A single‐pole low‐pass filter, consisting of an internal 9 kΩ resistor and an external user‐supplied capacitor is used to filter the output data and sets the pass bandwidth of the RATEOUT signal. Since these gyros are single axis devices, three devices are required to provide angular rate sensing in three planes X, Y and Z as shown in Figure 5(c).

The accelerometer chosen for this application was the KXM52 device from Kionix (2005). This is a high‐performance tri‐axial sensor that consists of a sensor element and an application‐specific integrated circuit packaged in a 5×5×1.8 mm package. The sensor element functions on the principle of differential capacitance, with device acceleration causing the displacement of an internal silicon structure resulting in a change in capacitance. The sensitivity of the device is factory programmable allowing customization for applications requiring ±1.0 to ±6.0 g ranges and for this application, the ±1.0 g device was chosen. As with the ADXRS150 gyroscope, the sensor bandwidth is user definable by selecting the value of the output filter capacitors.

The chosen MCU was the ATMega128L device from Atmel (Atmel, n.d). This device is a high‐performance, low‐power AVR^® eight‐bit MCU with an advanced reduced instruction set computer architecture with 53 programmable interface connector lines contained in a 64‐pin 10×10×10 mm thin quad flat pack package. The device also provides 8×10‐bit ADC channels that are used for sensing the gyroscope and accelerometer signals. Some other features that are used in this application are the power‐on reset and programmable brownout detection capabilities as well as an internally calibrated resistance‐capacitance oscillator. Also critical to low power operation are the sleep mode, power‐save, power‐down and standby modes available with this device.

The selected wireless transceiver was the NRF2401 form Nordic Semiconductor (n.d.). This is a single‐chip radio transceiver for the world wide 2.4‐2.5 GHz industrial, scientific and medical (ISM) band with a small package size of 5×5×0.7 mm. This transceiver features embedded protocol software to minimize power consumption and maximize data throughput using its Shockburst^® feature.

The chosen antenna for the application was a dielectrically‐loaded chip type (Antenna Factor, n.d.). The available form factor of 6.5×2.2×1 mm as well as the favourable electrical and radiation characteristics made it the most suitable commercial antenna for the WIMU application. These antennas use low temperature, co‐fired ceramic technology to allow the antenna size to be minimized. The electrical specifications include a 50 Ω characteristic impedance, omni‐directional radiation pattern, greater than unity gain, linear polarization and a usable bandwidth of 180 MHz in the 2.45 GHz ISM band.

D. RF subsystem

Having described the various system components, the RF subsystem is now described in more detail. In particular, the interface between the radio transceiver and antenna (or balun circuit) is described. The balun is a key component in balanced circuit topologies, typical of modern radio transceiver architectures. The term balun is an abbreviation of the words balance and unbalance and is defined as a device that converts a balanced system such as a two‐conductor line to an unbalanced coaxial line (Balanis, 1997). The balun plays an important role in maximizing radio communications performance while minimizing power consumption.

To explain the need for a balun, Figure 7 shows two cases in which a balanced dipole antenna is driven first by a balanced line and secondly via an unbalanced line.

In Figure 7(a), a current I₁ flows on the left conductor of the balanced line into the left leg of the dipole antenna while an equal but opposite current I₁ returns on the right conductor. Since both the currents are equal and opposite on the line, there is a net cancellation of currents on the line. In Figure 7(b), the balanced transmission line is replaced with a coaxial line. In this case, the currents I₁ and I₂ cancel each other within the line. However, since current I₁ cannot penetrate beyond the shield, a net current I₃ flows on the outer shield. The consequences are that the current in the shield becomes, undesirably, part of the radiating system and give rise to unwanted radiation from the line itself. The presence of this radiation from the line modifies the radiation pattern of the antenna and affects the impedance of the antenna. A balun is therefore required to suppress the unwanted current on the outer conductor of the unbalanced line.

Baluns are commonly realized by using either of three types of circuit elements: discrete transformers, transmission lines, or lumped element components (Balanis, 1997; Lee, 1998; Bakalski et al., 2002). Traditionally, baluns operating at microwave frequencies are designed using distributed topologies. However, at frequencies used for wireless sensor nodes such as the 2.45 GHz ISM band frequencies, distributed designs tend to be large and consume a large area on the PCB. For highly miniaturized wireless nodes, lumped‐element designs, consisting of discrete inductors, capacitors and resistors are popular due to the small size possible when surface mount technology components are used. The balun implementation for this application is shown in Figure 8.

The balun has four different tasks:

1. 1.

   implement a balun function by transforming the balanced transceiver signals (denoted Out_P and Out_N in Figure 8) to balanced antenna signal (denoted Ant_50R);

2. 2.

   transform the output impedance of the radio (100−j175 Ω) to an impedance of 50 Ω for the antenna;

3. 3.

   suppress the level of transmitter output harmonics to meet the required radio emission regulations; and

4. 4.

   suppress local oscillator (LO) signal leakage in receive mode.

E. Power subsystem

For this application, the WIMU is intended to be battery powered and as mentioned previously, the node needs to be power aware and keeping current consumption to a minimum is critical in order to maximize battery life. The power distribution circuitry has been designed to allow various circuit elements to be switched off under software control. Using this scheme, the MCU is powered continuously but in order to save power, it can be programmed to enter sleep mode or standby mode. The regulators were chosen to have enable signals under MCU control. Table I shows the power control functions of the main subsystem elements.

F. Interconnect design

The first step in specifying the interconnect subsystem was to determine the number of PCB layers or layer stack up needed to successfully implement the design. Ideally, a solid ground and supply plane would be required to provide a low‐impedance power distribution system for the sensitive analog sensor electronics as well as the noisier digital electronics. A good deal of time was spend in performing trial PCB layouts of the design and it was determined that the design could be implemented using a two‐layer stack up. Another factor that was considered at this time was the types of commercial flex material available and the required design rules for these types of materials. In consultation with various flex manufacturers, it was determined that utilizing a four or six layer stack up would mean that it would be extremely difficult to bend the substrate with a small radius of curvature, even for the thinnest flex material available. It was determined that a 90° bend would be possible with a four‐layer stack up but with a much larger bend radius of >1 mm. This was found to be unacceptable as such a large radius of curvature would add a further 2 mm to the outer dimensions of the folded cube and would add gaps along each folded edge, leading to a larger and weaker structure. After careful analysis, it was concluded to use a two‐layer stack up would be sufficient.

In order to realize a WIMU in a small cube structure, the choice of materials would be critical. A good deal of experience had already been gained from previous work using commercial thin‐film flexible polyimide substrate (Barton et al., 2007; Majeed et al., 2005, 2006). Here, a 3D module consisting of a thin, flexible polyimide substrate with folded flex was fabricated and tested. This work showed that folded flex provides a very good potential solution to the system packaging challenges inherent in implementing functional miniaturized sensor modules. For this particular implementation, the flex material would need to be flexible enough to allow a 90° fold to be made at the bend planes in order realize a cubic structure while at the same time ensuring that the flex substrate would not tear or crack. In addition, the substrate would need to have a certain degree of rigidity to ensure that after folding, the cube shape would be maintained before final assembly and packaging.

For this work, a commercial flex material AP‐8525 from Du Pont™ (Du Pont, n.d.) was investigated. This is a flexible, double‐sided, all‐polyimide copper clad material and has a good balance of electrical and mechanical properties especially bend and crease performance that is important for this application. In addition, the material has excellent thermal resistance, unique thick‐core product for controlled impedance and excellent dielectric thickness tolerance/electrical performance, all of which are very important properties for this RF application. Some of the key material properties for the flex material are shown in Table II.

In terms of choosing an appropriate soldermask material, conventional soldermask is not suitable due to its rigidity and tends to crack during bending.

Therefore, for this application, a flexible soldermask or “coverlay” was investigated used to ensure that the board could be bent without cracking. In particular, a Pyralux, LF0110 coverlay composite material (Du Pont, n.d.) was recommended by the PCB manufacturer. This material is a flexible substrate with and adhesive backing that is applied to the surface of the flex circuit to provide environmental and electrical insulation and some of the main properties are shown in Table III.

A flexible solder resist NPR‐80 (Nippon Polytech Corp., n.d.) was chosen designed to prevent solder bridges during soldering. In order to pattern the coverlay and soldermask, conventional lithography was discussed with the PCB manufacturer but it was determined that laser ablation is a better alternative to traditional lithography. In particular, laser ablation is easier to use on flexible substrates and produces more accurate features. Having determined the best flexible materials and manufacturing process for this application a decision was made to choose the manufacturer that could best meet the PCB layout design rules. One of the key design goals with respect to manufacturing was guaranteeing the reliability of the FPC once a 90° bend had been created. Some useful guidelines were provided by the chosen FPC manufacturer in order to meet this goal:

• •

  Build prototype system using an initial test substrate to allow characterization of the material for the application and then refine for final system.

• •

  The copper laminate thickness should be kept to a minimum.

• •

  Use top and bottom‐side coverlay braces at the bend planes to reduce the risk of interconnect failure during folding.

• •

  Only plate pad locations to minimize trace copper thickness and increase flexibility.

• •

  Pre‐crease circuits at bend planes to encourage the circuit to bend at the desired point.

• •

  Evaluate nickel‐gold finish for prototype and replace as necessary.

A. Prototype material specification

Following the manufacturer's guidelines, a prototype flex circuit was first specified, with the decision made to place additional coverlay strips at the bend planes as shown in Figure 9 to increase the strength of the substrate during bending. Table IV shows a summary of the stack up and materials used.

B. Prototype system fabrication and failure analysis

The prototype FPC was fabricated and the top and bottom surfaces of the finished substrate are shown in Figure 10(a).

On closer inspection, the manufactured plated‐through‐hole (PTH) via quality was discovered to be a problem as shown in Figure 10(b). A microsection analysis shown in Figure 10(c) confirm that breakout of the PTH vias had occurred. This was later identified to be caused by the manufacturer having drilled the holes too large. It can be seen that in some areas, the PTH vias had a minimum of approximately 8 μm of copper down the PTHs which was also identified as a reliability concern.

In addition to the above manufacturing defects, a bend test was performed on the prototype flex. In order to allow the subsequent folding of the substrate in an accurate and controlled manner, a dedicated test jig was designed specifically for this purpose. Figure 11 shows the jig of brass construction, with an accurate milled cut‐away to allow placement of the delicate substrate prior to bending.

Once the substrate was placed, the individual faces could then be accurately creased by carefully applying pressure at the bend plane with a beveled tool. A number of issues were identified post‐bending. Figure 12 shows cracking of the soldermask and metal close to one of the bend planes, just to the left of the coverlay material.

This was investigated further and was attributed to the presence of a large area of copper on the associated bend plane. This copper area was included to provide a good RF return path between the radio transceiver and antenna. The effect of the copper during bending was to move the stress at the bend plane further to the left towards the balun components and this would need to be addressed in the final system. In addition to the above tests on the unpopulated FPC circuit, a number of tests were also carried out on the populated FPC substrate shown in Figure 13.

Initial functional tests were performed by programming the WIMU modules in planar form. Then, having bent the substrate into a cube, an additional issue was observed in relation to the performance of the gyroscopes. This problem was eventually traced to buckling of the substrate at the corners of the centre face with open circuits forming on the ball‐grid‐array (BGA) gyroscope pins devices after bending as shown in Figure 14.

The bend planes on the centre face were surrounded by strips of additional coverlay material and it was determined experimentally that these areas were the most difficult to bend. In addition, open circuits were observed in areas where component pads were placed near the bend plane as in Figure 14(c). This problem manifested itself on the traces with minimum width (137 μm) and did not occur on the wider traces or traces that did not have coverlay apertures in the coverlay as in Figure 14(d). This effect was also seen in a previous work (Barton et al., 2007) where coverlay openings on the bend plane exhibited metal cracking after bending. Of six circuits that were fabricated and tested, five failed during bending, resulting in a failure rate of approximately 83 per cent. Obviously, this failure rate and unreliability is unacceptable for the final implementation. Having identified a number of serious issues with the prototype flex circuit, a number of corrective steps were identified for the final circuit implementation as follows:

1. 1.

   Use of a thinner substrate material and alternative finish to nickel‐gold.

2. 2.

   Eliminate use of coverlay at bend planes and instead use a thin flexible soldermask for the entire flex circuit.

3. 3.

   Eliminate copper areas overlapping at bend planes.

4. 4.

   Include 1 mm holes at all bend planes to avoid creasing/buckling at corners.

5. 5.

   Widen all pad entry traces close to the bend plane to reduce risk of open circuits.

6. 6.

   Pre‐fix BGA devices prior to bending.

The finish for this build was also specified on this occasion as an immersion tin finish which is more flexible than the previous finish of nickel‐gold. Figure 15 shows a cross‐section of a PTH via for the finished substrate. It can be seen that via formation is now very much improved, with the drill oversize problem of the previous revision having been corrected and a minimum of approximately 20 μm for the via side walls.

In order to characterize the new material, the same bend tests as described previously were carried out. It was found that the new material was much more flexible and did not have the same issues during bending. The addition of the corner holes reduced the stresses that were evident previously and the removal of the copper at the bend plane between antenna and radio transceiver eliminated the cracking problems. Based on the positive results obtained, the thinner flex material was chosen for the final fabrication of the WIMU.

C. Final system fabrication and failure analysis

Prior to population of the final flex circuit, a bend was formed on each bend plane and then bent back in the opposite direction. This left a flat crease along each bend plane, making it easier to fold the circuit once populated. The pre‐creased PCBs were populated with components and the final planar PCB assemblies are shown in Figure 16(a). In this configuration, the assembly was programmed and tested prior to folding into a cube form factor. Once tested, the gyroscopes were secured with epoxy prior to bending to avoid the risk of the BGA gyro pins lifting and causing open circuits after folding. The assemblies were then folded and once the cube shape was formed, the adjacent edges of the cube were fastened to each other using an epoxy resin. The final folded cube is shown in Figure 16(b).

Of the ten final WIMU systems that were characterized and tested with the new substrate material, a significant improvement was observed with no failures being recorded after folding the flex into a 3D cube.

A. Software development

The embedded software on the WIMU transmitter and receiver are now described. The transmitter software consists of various functions that perform a number of tasks in order for capture of the gyro and accelerometer measurement data and the wireless transmission of this sensor data. The radio transmits each byte of data on a chosen channel (1) of the available ISM bandwidth between 2.4 and 2.48 GHz. On the receiver side, the embedded software on the WIMU receiver consists of various functions that capture the transmitted gyro and accelerometer measurement data from the WIMU transmitter. A user‐friendly GUI was developed using Labview in order to display the captured motion data from the WIMU transmitter. An example of the GUI output screen is shown in Figure 17, displaying the measured gyroscope and accelerometer outputs in real time when the WIMU is moved.

B. System calibration and test

Since the WIMU module has been developed using off the shelf components, the performance of the module had to be tested for basic functionality and then for the generation of a set of system specifications. A number of experiments were designed and executed to carry out this characterization. The system was connected to a PC wirelessly to log the information received from the WIMU for off board analysis. Each of the sensors was subject to conditions that excited the respective sensing properties (tilt, rotation, etc.) and the results were recorded and compared to the manufacturers' specifications. The accelerometers were subject to tilt extremes to measure the max, min and zero offsets. The gyroscopes were excited using a calibrated rotating table. All of the data were recorded and analyzed used a customized Labview application. As described previously, the Labview GUI application was developed to provide a visual analysis of the sensor data being produced by the WIMU. This application allowed for the viewing of the sensor signals in real time and contained integrated calibration routines that allowed for the calculation of the resolution of each the sensors.

In order to ensure that the WIMU is calibrated properly the voltage offset, sensitivity and non‐orthogonality of the gyroscopes and accelerometers need to be evaluated so their effects can be mitigated during performing measurements. In addition, for accurate calibration, the mutual misalignment of the sensors to each other needs to be determined. However, determining the errors due to sensor non‐orthogonality and mutual misalignment requires complex mathematical methods that are not treated here. In this work, a relatively simple calibration method for estimating offset and sensitivities is used while the non‐orthogonality and alignment issues will be the aim of future work. For calibration, measurements made directly from the IMU are referenced with respect to the three defined IMU sensors axes as shown in Figure 18.

In order to characterize the accelerometer offsets and sensitivities, it is required to know the maximum and minimum values of acceleration when the accelerometer is oriented in two specific configurations. The first of these configurations is with the accelerometer held stationary and one of its three axes pointing towards the earth's core. In this case, the sensor output will be a maximum while experiencing a force of 1 g or 9.81 m/s² and the sensor output voltage is denoted here by the value V_max. The second configuration is with the accelerometer's sensor axis perpendicular to the earth's surface but pointing away from the earth's core. In this configuration, the sensor output will be a minimum, denoted by the value V_min. The measured acceleration offset is then given by equation (1), where: Equation 1 The measured accelerometer sensitivity is given by equation (2), where: Equation 2 In order to estimate the voltage offset of the gyroscopes, it is necessary to rotate the WIMU about the three principal axes and then calculate the average gyroscope output for each orientation. The estimation of the gyroscopes sensitivities was carried out using a digital turntable. The WIMU module was placed on the turntable ensuring the table's axis of rotation was in line with the sensor's sensitive axis (note that the angular velocity will be the same for all distances from the turntable's centre). The table was set to a nominal angular velocity of 20 rpm. The received data were logged and an example of the gyroscope X‐axis output from this calibration is shown in Figure 19.

The measured angular velocity is very accurate for angular velocities less than 30 rpm and is limited by the sensitivity of the gyro. This is not an issue though as the sensitivity of the gyro is easily changed by adjusting the value of a single resistor, depending on the application. The complete characterization and calibration measurements are shown in Table VI.

The measured calibration constants for each sensor were then incorporated into the WIMU firmware and GUI application to ensure the accuracy of subsequent system measurements.

C. RF testing

Some simple tests were carried out in order to test the RF communications range of the WIMU. The antenna performance was tested in an open‐field environment, free from nearby obstructions that could affect the accuracy of the measurements. The transmitting mote was configured to send 100 packets of data for subsequent reception by the receiving mote. The transmitter and receiver motes were each mounted on a wooden pillar to a height of 1.5 m. The packet loss was then measured as a function of distance by varying the horizontal separation between the transmitting and receiving motes. This was accomplished by moving the transmitter while leaving the receiver stationary and measuring the packet loss for each measurement. A laptop was connected to the receiving mote via an RS232 cable in order to capture and display the results that are shown in Figure 20.

It can be seen that WIMU system has a range of 60 m approximately in open‐air. The module was tested indoors and the range fell to 20 m approximately in a typical Lab‐corridor environment. This is expected and is attributed to the multi‐path fading effects that are experienced in non‐ideal RF indoor environments.