New generation of High Performance/High reliability MEMS accelerometers for harsh Environment

F.Rudolf, S.Gonseth, R.Brisson, P.Krebs Colibrys (Switzerland) Ltd

Av. des Sciences 13 1400 Yverdon-les-Bains, Switzerland

felix.rudolf@colibrys.com

Abstract— This work aims to push the limits of MEMS open

loop accelerometers towards higher performance, higher

reliability in increasingly harsh environment. The new generation

accelerometer relies on proven bulk micromachined capacitive

MEMS sensor and a new ASIC. The aim is to improve noise,

linearity while being able to operate reliably under strong

radiation and at high temperature. A new ASIC was developed

and tested. It has a noise of 1.7 ppmFS/root Herz, nonlinearity

smaller than 0.2 %, operates up to 175°C. Radiation tests have

shown that the analog part of the ASIC operates well for total

dose of 57 krad and under heavy ion bombardment.

Keywords—MEMS accelerometer, rad-hard ASIC, high

temperature accelerometer

I. INTRODUCTION

Many applications require accelerometers to provide precise and reliable data even under harsh environment and still to be small and consume low power. Such applications are:

• Measurement of tilt and vibration in drilling tools for the oil and gas industry, requiring operation at high temperature and in the presence of heavy vibration and shock.

• Flight and robotics control on space missions requiring reliable operation even after and during extended exposure to radiation.

• Various applications such as personal navigators, automotive testing and train monitoring where accelerometers are exposed to high repetitive shocks (e.g. devices mounted on shoes or on car chassis) where accuracy, size, power and robustness matters.

The aim of this work is to develop an accelerometer to address these needs. It has already been demonstrated that bulk micromachined capacitive MEMS accelerometers are very robust [1,2]. They even can withstand the firing shocks of 20’000 g when used in guided munitions. The limiting factor for many applications is the electronics transforming the capacitive signal into an electrical output. Previous electronics used was not designed for high temperature nor for immunity against radiation effects such as radiation induced upsets SEU/ MBU and large total doses. There are basically two ways of measuring the acceleration with a MEMS sensor: either open or closed loop. In an open loop electronics the capacitance change is the MEMS is measured and amplified. In closed loop

electronics the inertial forces are compensated by electrostatic forces [3]. Closed loop system allows reaching better ultimate performance in terms of bias stability, linearity and noise. The price to be paid for these ultimate performances is in terms of power (needs very precise high voltage), size (driven by the power supply requirements), and complexity (analogue and digital electronics). For this work open loop electronics has been chosen. As compared to closed loop systems it has a 10 fold advantage in power, size and is significantly less complex, while still reaching the performance required for the target applications.

II. DESIGN

Key requirements for high performance and high reliability

in harsh environments are a stable mechanical sensor that is a MEMS device and its associated die attach technology.

For the MEMS sensor a proven capacitive accelerometer design is used as illustrated in Fig.1. It is based on a proof mass suspended by a spring and detecting acceleration in the out of plane direction. It has already shown that excellent performance can be reached with this approach [2].

Fig 1: MEMS capacitive accelerometer cross-section. An out of plane acceleration will deflect the proof mass and change the capacitances between the middle and the top and bottom plate respectively.

A new die attach technology was developed that is robust in high repetitive but has still minimal die attach stress.

The second key element is the design of the electronics for operation over in all environmental conditions, i.e. over the full temperature range and under radiation. In a first phase an ASIC

was designed for industrial applications, including operation at high temperature (175°C). The bloc schematics of the new ASIC is shown in fig. 2. The capacitive signal from the MEMS sensor is measured by the differential charge balancing loop bloc. In a control loop the capacitor bridge is balanced. This concept [4] has already been successfully used in the previous ASIC. In order to improve stability and noise, a fully differential configuration is used providing a positive and negative output signal, the acceleration signal being the difference between the two signals.

Differential

charge balancing

loop

Gain & offset

Sequencer block

T°Csensor

RegistersOTP

Serial interface MISO

SS

Output +

Output -

T

SCLK

MOSI

VDDVSS

POR

Oscillator

Input Input / outputOutput

ASIC

ST

CLK_ext

Reset

lowpass filter

Err

Cs1

Cs2

Fig 2. Bloc schematics of the ASIC

The charge balancing loop includes also features to adjust the linearity caused by stray capacitances and sensor nonlinearity. A significant advantage of this concept is that the electrostatic forces applied on either side of the proof mass are always equal, independently of the plate position this reduces significantly measurement errors caused by electrostatic forces especially for low g accelerometers. The signal form the charge balancing loop is the fed through a programmable amplifier (PGA) and a programmable low pass filter, to provide an easy to use output voltage to the user. The differential output assures that zero g corresponds to positive and negative output being equal, independently of any reference voltage. The gain is ratiometric to the power supply voltage i.e. if the power supply is used as a reference for the subsequent A/D converter the accelerometer output becomes largely independent of any power supply voltage variation.

The service blocs include a programmable memory (OTP) to store the calibration data, an RC oscillator and power management blocs. The digital interface is only used in manufacturing, i.e. to write the calibration data in the OTP memories. The ASIC is designed to operate between -55°C and 175°C to full performance.

The ASIC includes a temperature sensor that can be used for temperature correction by the user.

The ASIC operates out of 3.3V, and uses less than 10 mW.

A radiation hard version of the ASIC is under development. Based on the results described in chapter III, it is designed for immunity against radiation as well for single events as for a total dose of 100 krad. The critical part for radiation hardness is the digital blocs needed for calibration. A triple redundancy voting system is used to check the digital information. In case of a single or multiple set-ups (SEU / MBU), the data are

refreshed from the OTP’s to correct the errors. This is done in the background, without affecting the operation of the analogue signal chain.

The ASIC and the MEMS are packaged in a hermetically sealed ceramics package

The design values for the accelerometers are shown in table 1. As this accelerometer design covers full scale ranges between ±1 g and ±100 g the design values are given with respect to full scale mostly in ppmFS.

Table 1 Design values of the accelerometer.

Working ranges 1 to 100 g Bandwidth LPF 500 and

2000 Hz

Performance Bias stability 50 ppm FS Scale factor Nonlinearity

0.2 % FS

White noise 0.6 ppmFS/√Hz Electrical Supply voltage 3.3 V Power < 10 mW Environment Temperature range

-55 to 175 °C

Radiation* 100 krad Shock survival 10’000 g

*Radiation hardened version of the ASIC

III. ACCELEROMETER MANUFACTURING AND TESTING

Prototypes of the accelerometers and the ASIC in a non-rad hardened version where manufactured. The accelerometer is shown in Fig 3.

Fig. 3. Complete accelerometer with the MEMS (left) and the ASIC (right) in a hermetically sealed ceramics housing.

A new die attach technology was applied to make them very robust. This technology is already applied in the existing

Colibrys RS9000 product family. Shock test were performed on the RS9010 devices to qualify the technology.

Two tests were performed, one with multiple shocks of low amplitude and one with multiple shocks of high amplitude.

The low amplitude shock test consisted of 20 shocks with 740g. These test were performed at -30, 20, 80°C. The measured bias shift was < 44 ppmFS. This die attach technology is also applied for the new generation of accelerometers

The high shock test consisted of 90 shocks of 6000g, 0.15 ms. Tests were performed on 13 samples. No sample failed. Figure 4 shows the variation of bias during these tests. The maximum observed bias shift was 1250 ppmFS and the sensitivity shift > 400 ppm. As a comparison, devices with the previous die attach technology would generally fail under these conditions.

Fig. 4. Evolution of bias during the multiple shock tests

Accelerometer prototypes with the new electronics were calibrated and in depth qualified with static and dynamic measurements.

A critical step is the nonlinearity calibration. The charge balancing loop has two parameters for adjusting the even and odd nonlinearity components independently. A calibration method based on dynamic excitation and on the nonlinearity measurement described in [5] is used. This method is based on a fitting of the response curve with a 6th order polynomial and subtracting the constant term (bias) and the first order term (sensitivity). The remaining terms represent the nonlinearity. For a given curve this method gives factor 2 to 3 worse results than the often used deviation from a straight line fit. This has to be kept in mind when comparing different nonlinearity data. The calibration is done against a piezoelectric reference accelerometer. This is clearly a limitation since the nonlinearity of the reference accelerometer is of the same order as the nonlinearity goal of the accelerometer to be tested. Typical

nonlinearity curves of 9 calibrated accelerometers are shown in Fig 5.

Fig. 5. Nonlinearity of 9 calibrated 30 g FS sensors

The long term behavior was assessed by a measuring the bias shift over a period of > 55 hours powered at 85°C. This corresponds roughly to 2 years of lifetime at room temperature

assuming electronics standard activation energy of 0.8 eV. This measurements shows that bias shift due to ageing is less than 100 ppmFS. A typical result is shown in fig.6.

Fig. 6. Bias drift measured at 85°C over > 55hours. The 4 curves are from 4 different devices.

Fig. 7 shows the temperature dependency of bias. Preliminary assessment of the temperature behavior. The curves represents data points measured over two temperature cycles 20 � 80 � -30 � 20°C. The offset temperature coefficient measured on 10 samples is typically - 9 ppmFS/°C. The deviation from a straight line fit is < 60 ppmFS. This represents the maximum repeatability error caused by temperature hysteresis and possibly measurement errors.

Fig. 7. Bias shift over temperature

The noise spectrum is shown in figure 8. The white noise is 1.7 ppmFS/√Hz and the Flicker noise corner at 0.2 Hz. This value higher than the target value of 0.6 ppmFS/√Hz. An in depth analysis of the circuit has revealed the dominant noise sources that will be corrected in the final ASIC design.

Fig. 8. Noise spectrum. The data was acquired at 200 Hz with a filter at 90 Hz. The dotted lines show the models for Flicker and white noise

Fig. 9 shows an Allan Variance. It is in good agreement with the noise measurement. The Allan Variance shows that the bias drift due to Flicker noise is below 1ppmFS.

Figure 9: Measured Allan Variance

IV. RADIATION HARDENING

In order to assess the gap to be closed between the industrial ASIC and a radiation hardened ASIC a set of tests were performed. Several test structures and an early prototype of the new ASIC not designed for radiation hardness were tested under gamma radiation up to a total dose of 100 krad and were operated under heavy ion bombardment. The goal of the total dose test was to identify any drifts induced by the radiation. Measurement under heavy ion bombardment allowed to test latch-up immunity of the design and to evaluate the effect of radiation events on the output signal.

In the total dose experiment the sensor characteristics were measured before and after radiation. The ASIC was subject to Gamma radiation (Cobalt-60 source) with a total cumulated dose of 54 krad. A potential effect of radiation could be a change of the characteristics of the capacitor bank used for non-linearity compensation.

Fig. 10 shows that even a heavy total dose did not induce changes of the non-linearity compensation. In this set-up the nonlinearity calibration was not optimized and the measurement accuracy was not sufficient, hence the spread of some of the data points. For this nonlinearity measurement the straight line fit method was used.

Fig 10. Nonlinearity measured before, after radiation and

after an annealing step In the heavy ion bombardment experiment the sensors

were operated during irradiation with 150 MeV Xe ions at fluence of 106 Ions/cm2. The output data were recorded at high speed without filtering.

On the output signal there were short spikes with 10 to 40 mV amplitude and 10 µs duration. They are illustrated in Fig. 11.

Fig 11. Spikes observed on the output signal of the ASIC

the top and bottom curves are the positive and negative channel of the differential charge balancing loop bloc.

These spikes will appear as signal on the accelerometer output. However since they are so short, they are almost completely filtered when passing through the 500 Hz low pass filter of the ASIC. These tests showed that the ASIC and the chosen technology are immune against radiation induced latch-up. The test circuit included also different digital blocs for calibration purposes. On these digital blocs many SEU and possibly MBU events were seen leading to changes in the sensor calibration. These test showed that the analogue part was already immune against radiation effects but that the digital blocs need radiation hardening strategies to make the circuit layout radiation hard and the system design error tolerant

V. DISCUSSIONS AND CONCLUSIONS

This accelerometers show a significant performance improvement as compared to state of the art products.

Noise: To compare the noise of different accelerometers one can use the white noise density normalized to the full scale acceleration range. It can be expressed in terms of ppm FS/√Hz. State of the art low noise open loop silicon MEMS accelerometers available on the market have noise down to 8 ppm FS/√Hz. Thanks to a consequent design for low noise and

the relatively large sensor capacitors a noise level of 1.7 ppm FS/√Hz could be demonstrated, that is a 13 dB improvement. Analysis has shown that the noise limit of this concept can even be further lowered to 0.6 ppm/FS. Only closed loop MEMS accelerometers such used for seismic imaging have lower values in the range of 0.2 ppmFS/√Hz.

Nonlinearity is especially important when low frequency acceleration is to be measured in a vibration environment. Vibration leads to vibration rectification errors due to nonlinearity. Therefore nonlinearity is a key parameter in harsh environment. In this typical nonlinearity is 0.05% with a worst case of 0.2%. This has to be compared to state of the art MEMS accelerometers that have typical Nonlinearities between 0.2 % and 0.5% but with maximal values that are often in the 1% range.

The good basic performance combined with the ability to operate in harsh environment such as radiation and high temperatures. This makes this accelerometer well suited for applications for guidance applications in space vehicles (e.g. descent and landing operations and rovers), measurement of tilt and vibration in drilling tools for the oil and gas industry and vibration monitoring in exposed applications such as car chassis or train bogies. The low noise will also allow addressing many structural health and earthquake monitoring applications.

REFERENCES

[1] J.-M. Stauffer, MEMS Accelerometers: From Market to Advanced

Applications. Plans 2004, April 26-29, 2004, Monterey, California,

[2] J-M. Stauffer, O.Dietrich, B. Dutoit, RS9000, a Novel MEMS Accelerometer Family for Mil/Aerospace and Safety Critical Applications, IEEE/ION Position Location and Navigation Symposium (PLANS), Mai 4-6 2010, Indian Wells, California, USA

[3] P. Zwahlen, Y. Dong, A-M. Nguyen, F. Rudolf, P. Ullah, V. Ragot, High Performance Inertial Navigation Grade Sigma-Delta, Inertial Sensors and Systems 2012, September 18- 19, 2012 Karlsruhe, Germany

[4] R. Le Reverend, “Device for measuring a force with the aid of a capacitive sensor using charge transfer”, US patent US5821421

[5] IEEE std 1293-1998