INSTITUTE OF PHYSICS PUBLISHING J. Micromech. Microeng.

15 (2005) S109S115

JOURNAL OF MICROMECHANICS AND MICROENGINEERING

doi:10.1088/0960-1317/15/7/016

Design and performance analysis of thermally actuated MEMS circuit breakers

E M Yeatman
Department of Electrical and Electronic Engineering, Imperial College London, SW7 2AZ, UK and Microsaic Systems Ltd, GMS House, Boundary Road, Woking, Surrey, GU21 5BX, UK E-mail: e.yeatman@imperial.ac.uk

Received 13 December 2004, in nal form 19 April 2005 Published 20 June 2005 Online at stacks.iop.org/JMM/15/S109 Abstract The use of MEMS electrical switches as miniature electrically resettable circuit breakers is described. A prototype based on nickel thermal actuators on silicon has been fabricated and tested, and thermo-mechanical analysis is presented showing the relation between design parameters and performance. Temperature dependence of trip current is shown to be below 0.2% K1, and analysis is presented indicating that variation of resistivity is the main cause. Trip time, and its dependence on over-current level, is analysed and the results compared with measurements.

1. Introduction
Electrical switches fabricated based on MEMS technologies have been reported by a number of research groups, and while many of these are intended for high-frequency applications, switches suitable for dc and low frequencies have also been reported [1, 2]. DC switches inevitably require direct physical contact, and a variety of congurations are possible, with specic designs choosing between lateral (in-plane) and vertical movements, and between thermal, electrostatic or other forms of actuation. An attractive application for MEMS dc switches is in circuit protection, effectively to act as ultra-miniature electro-mechanical circuit breakers. The MEMS approach provides cost, size, functionality and integration benets over conventional relay-type circuit breakers. Since the maximum currents are likely to be a few amps at most, applications in electronics (e.g. on-board over-current protection) are the most suitable. Although fuses are often used in such applications, increasingly there is a requirement for recoverable protection without component replacement [3]. Therefore the conventional fuse is not the main technology against which MEMS devices should be compared; rather, solid-state electronic devices, and positive temperature coefcient (PTC) devices, are the main alternatives. The former, typically silicon MOSFET switches, are convenient
0960-1317/05/070109+07$30.00

for integration but tend to trip too readily, consume power, and can fail in the short circuit state, so these are far from ideal as fuse replacements. The PTC is small, low cost, and recovers its conductive state after the over-current condition ends. However, it has serious drawbacks that a MEMS solution can overcome: slow and variable trip times, high ambient temperature dependence and lack of controllability of resetting. We have recently demonstrated working prototypes of MEMS circuit breakers [4], and provide here a detailed analysis of their design and performance. While MEMS thermal actuators have been analysed by a number of authors, circuit protection has different requirements from other applications. In this paper, two issues of particular importance to circuit breakers are analysed in detail: ambient temperature dependence and variation of trip speed with drive current.

2. Device structure
The device we have developed is based on latching, laterally moving thermal actuators. The actuators are of the shape bimorph (Guckel) variety [5]. Figure 1 shows a schematic of the device, with the set and trip actuators, contact points and external terminals indicated. When the actuators are latched together, the load current passes from the source connected at D to the load connected at E, either across the latch contacts S109

2005 IOP Publishing Ltd Printed in the UK

E M Yeatman

contacts
E

hot arm D C latch set actuator cold arm trip actuator

L c L hs Les

Le

A B

Figure 1. Schematic of MEMS circuit breaker. Figure 4. Dimensions for analysis of set actuator.
oxide Si Ni Au

of the hot arm. Since the latch trips with minimal deection of the trip actuators, the actuator stiffness can be neglected, so equating moments we arrive at a required Fn = s FL (Le /g ). This is obtained from the thermal expansion of the hot arm according to A Lh (T (x) TA ) dx (1) Lh 0 where E and are the Youngs modulus and thermal expansion coefcient of the actuator material, A is the hot arm crosssectional area, and T(x) TA is the temperature distribution with respect to the ambient. The hot arm expansion results from Joule heating caused by the load current I passing through the hot arm. If we assume that the cooling is dominated by conduction down the arm to its two ends, and that these remain at ambient temperature, then the one-dimensional thermal equation is easily solved giving a quadratic variation of over-temperature: Fn = E I 2 (Lh x x 2 ) (2) 2A2 k where and k are the electrical resistivity and thermal conductance, respectively. Note that in this initial analysis we have neglected temperature dependence of the parameters; this will be addressed in section 4. The integral of the over2 temperature given by (2) gives I 2 L2 h (12A k), which we can insert in (1) to derive the trip current: 1 E 2 gI R = s FL (3) 12 k with s the coefcient of static friction, the hot arm electrical resistance R = Lh/A, and approximating Le = Lh . Let us now consider the set actuator in detail, as illustrated in gure 4. In this case the actuator displacement is signicant, but we simplify the analysis by assuming the cold arm rotates an angle around the narrow hinge at its base, in reaction to the expansion of the hot arm. Then the stiffness of the whole actuator structure (including the hot arm) can be modelled as a linear rotational spring of stiffness Ka, such that to rotate the undriven actuator by , a moment M = Ka must be applied. When driven by a set current, the hot arm will expand unless restrained by an external force; in response to such a force it will generate a moment Ma which will fall linearly from a maximum value Mao at = 0, to zero at max, where max T (x) TA =

Figure 2. Cross section of device structure.

Fe FL

wh

Lh

Fn

Le
Figure 3. Dimensions and forces for analysis of trip actuator.

or through the cold arms of both actuators via an external connection between C and B. When the load current exceeds the rated value, the trip actuator produces sufcient force to trip the latch, and the contact is broken. Resetting is achieved by applying a reset current between the set actuator terminals A and B. The device was fabricated using nickel electroplated actuators on a silicon substrate, with gold alloy contact metallizations [4]. The overall device structure is illustrated in gure 2.

3. Actuation analysis
The required trip force depends on the latching force, which can widely be varied by adjusting the stiffness of the set actuator and the displacement needed to set the latch. However, the latching force FL will strongly be related to the contact force Fc , which in turn will set the contact resistance. The optimal FL will be the minimum that achieves sufciently low contact resistance, so as to minimize the trip actuator size. Previous studies suggest a few hundred N as suitable contact force values [6]. A detailed analysis of the set actuator (below) gives the relation between FL and Fc . First the tripping performance is analysed with reference to the geometric parameters indicated in gure 3. The latching force FL results in a static friction force sFL which must be overcome by an equal force Fe. This force is in turn provided by a normal force Fn which results from heating S110

Design and performance analysis of thermally actuated MEMS circuit breakers

is the angle reached in the unloaded state, with a slope Ka. Note that if we neglect the hinge stiffness, and use a small angle approximation, then max = L /gs, where L is the unloaded extension of the hot arm when driven, i.e. L =
0 Lhs

1 Kc /K a = 10 0.8 2

(T (x) TA ) dx.

(4)
Fc /Fc(max)
0.6

As the set actuator rotates in response to its drive current, it reaches an angle c at which the contacts close. For further rotation, the contact force Fc will induce a moment Fc Lc , which can also be written Kc( c), with Kc the additional rotational stiffness induced by the contact support. This stiffness can be adjusted independently of Ka in the design, by varying the width of the extension arm on which the moving contact is mounted. If we neglect the force applied by the latch parts as they are pushed apart (this can be made quite small using long latch arms), then in the driven state (set current on) we have: Mao Ka = Kc ( c ). (5)

1 0.4 0.5 0.2 0.1 0 0 0.2 0.4 0.6 0.8 1

c /max

Figure 5. Variation of contact force with contact closing angle, for various stiffness ratios.

If the set current is chosen so that the latch just engages, then the solution to (5) will be the angle when the device is latched, L. Thus Mao + Kc c L = . (6) Ka + Kc If we cut the set current so that no actuation force is applied, the angle will stay the same (neglecting any backlash in the latch), so that the moments from the contacts and the actuator stiffness are unchanged, and thus the lost bending moment from actuation must be exactly compensated by that due to the latch force, i.e. Mao = FL Les . (7)

be well separated. How to achieve this is illustrated in gure 5, where Fc /Fc(max) is plotted versus c/ max for various stiffness ratios Kc/Ka. It can be seen that Kc/Ka 1 is needed to obtain well-spaced values of the three angles. We can also now consider the relative magnitudes of the contact and latch forces, which we require in order to design the trip actuator. Combining (7) and (8), and noting that Mao = Ka max, we obtain Les Ka Fc = FL Lc K c 1 L max . (11)

We can now obtain a number of relations between the three angles c, L and max, the stiffnesses Ka and Kc, and the forces Fc and FL, by manipulation of (5)(7) and using the associated denitions, as follows: Ka (max L ) Lc 1 Ka Kc Fc = (max c ) Lc K a + K c Mao + Kc c L = . Ka + Kc Fc = Practical considerations include the following: the gap between the contacts must be substantial when the actuator is tripped, to maximize the voltage that can be sustained without breakdown, so c should be as large as possible; L must be signicantly greater than c to obtain a strong contact force unless Kc is very high; however, if a high Fc is obtained with a small ( L c) using a high Kc, then Fc will be too sensitive to backlash, contact wear and other factors affecting the geometry; from (8) it can be seen that we require L < max in order to obtain any contact force. From these considerations we nd that within a certain max as constrained by maximum set power, c and L should (8) (9) (10)

We have seen above that suitable values for Ka/Kc and L/ max are 1 and 0.7 respectively, for which Fc /FL 0.3(Les/Lc). For a length ratio Les/Lc = 0.7, we have FL = 5Fc . Then for a contact force of 200 N, we require 1 mN latch force. We can now return to the trip actuator design. For nickel the combined constant E/k = 28 000 s m2 [4]. For g = 20 m, s = 0.8 [7], a latching force of 1 mN and a target hot arm resistance of 0.1 , we obtain a useful nominal trip current of Io = 0.4 A, close to what was measured (see below). Note that the hot arm resistance and the contact resistance act as series loads on the supply during normal operation, and so must be minimized. For a total series resistance of 0.2 including the contacts, a continuous load current of, say, 0.3 A will dissipate 18 mW. The fabricated devices had series resistances in the range of 0.11 in about 50% of working devices, and trip currents near the design value at 300400 mA [4]. Figure 6 shows a prototype device in the set (latched) state. Hot arm lengths were 2 mm for set and trip actuators, and maximum displacements at the latch parts were 60 m, giving max = 1.7 . With c = 1 , this provided a contact gap of about 30 m, as can be seen in gure 6. According to Paschens curve for air, this gap can sustain about 100 V. The hot arm widths and depths were 20 m and 60 m respectively. From (2) we can derive the approximate peak temperature (taking 2 x = Lh/2) as Tmax = I 2 L2 h 8A k . For the dimensions above, and taking for the thermal conductance and resistivity of nickel 90 W m1 K1 and 7 108 m respectively [8], we obtain a reasonable value of Tmax = 43 C. S111

E M Yeatman

(a)

(b)

Figure 6. Optical micrograph of circuit breaker (a), and close-up of latch area and contact in open state (b). Length of contact surface is 400 m.
actuation current (mA)
350 300 250 200 -10

10

30 TA ( C)

50

70

Figure 7. Trip current versus ambient temperature.

It is worth also considering the out-of-plane behaviour of the device. The structures are intended to move only in the plane of the substrate, but a number of effects could cause out-of-plane motion, including vibration and shock, stress gradients in fabrication, or temperature effects caused by expansion coefcient differences. To minimize these, a high aspect ratio was chosen (structure depth is 3 hinge width) so as to provide a high stiffness out-of-plane. Although there is a substantial expansion coefcient difference between the nickel actuators and the silicon substrate, the fact that the actuators are only anchored at one end avoids buckling due to temperature change. It can also be noted that the very low mass of the actuators limits their sensitivity to shock. The total mass of one actuator is 0.3 mg, so that for example, a 50 g shock load gives a force of only 0.2 mN, signicantly below the design latching force of 1 mN.

One possible explanation is the higher thermal expansion coefcient of the nickel actuators compared to the Si substrate. As the temperature rises, the trip actuator elongates with respect to the position of the set actuator (gure 5). The displacement needed to latch the two actuators together is therefore reduced. This lowers the holding force, and therefore the current needed to release it. If we dene = Ni Si, then for an ambient temperature shift T the latch will move to the left a distance Le T. This will rotate the set actuator by = Le T/Les. This will reduce the latch force and consequently the required trip current. It is straightforward to show that dFL/d = (Ka + Kc)/Les. From (3) we can see that the required trip current Io squared is proportional to the latch force, so dIo/Io = 1/2(dFL/FL). Combining (8) and (11), and approximating Ka = Kc and Les = Lhs, we can obtain (dFL/d )/FL = 4/(3 L). Then (dIo/d )/Io = 2/(3 L), and dIo /dT 2 Le . (12) = Io 3 Les e From [4], = 9 106 K1, and in the fabricated devices L = 0.02 rad and Le = Les. This gives a temperature sensitivity using (12) of 3 104 K1, or 0.03% K1. This is about 5 less than that measured, indicating that differential expansion of the substrate is not the main cause of temperature dependence. Another possible cause is the temperature dependence of the material parameters. Of these, the one with the strongest temperature dependence for nickel is the resistivity . To analyse the effect of this dependence, it is necessary to redo the analysis leading to (3) with (T ) taken into account. Fortunately, a linear variation of resistivity provides a good approximation for nickel over the range of 3090 C [9], which should cover the range reached within the hot arms in the experiments conducted. We dene = o + (T To ) (13) with o the resistivity measured at a reference temperature To. The temperature variation for Ni is usually taken as 0.6% K1;

4. Temperature dependence
Since the deection of the thermal actuators depends on temperature difference between hot and cold arms, rather than on absolute temperatures (as for PTC devices), we should expect low sensitivity of trip current to ambient temperature. The variation of trip current with ambient temperature was measured, as shown in gure 7, and the slope is indeed low at 0.15% K1 (about 10 less than for PTC devices). Since this is potentially an important advantage of the technology, it is useful to analyse what the main source of this dependence is. S112

Design and performance analysis of thermally actuated MEMS circuit breakers

T(x) - TA (deg.C)

for o = 7 108 m as above, this gives = 4.2 1010 m K1. The thermal analysis of the trip actuator including this effect is too lengthy for inclusion here, so instead we present the key results. An analytic solution for the spatial temperature prole can be obtained, which has a cosine form. Integrating this, we can derive Lh =
Lh /2 Lh /2

60 50 40 30 20

I = 0.4 A

I = 0.3 A

I = 0.2 A
10 0 -0.5 -0.3 -0.1 0.1 0.3 0.5

(T (x) TA ) dx 0 tan( ) 1 (14) is given by

x/L h

= Lh TA To +

where TA is the actual ambient temperature and =

Figure 8. Modelled excess temperature versus position on hot arm, for currents as indicated, with (solid lines) and without (dashed lines) temperature dependence of resistivity.
100

I Lh . (15) A k 2 A relation between the parameter and the peak temperature can also be found as follows: 2 A Tmax = Tmax TA = (16) 2 from which it follows directly that ( max A) = 2/2, with A the resistivity at TA, and max is the peak resistivity (at the centre of the hot arm). For the parameters of the devices fabricated, the value of is =0.7, so that a series expansion of tan ( ) in (14) is reasonable, but several terms should be taken. Taking three terms for tan gives tan ( )/ 1 = 2/3 + 4 2 /15, from which we obtain for the hot arm expansion: Lh = 1 A I 2 L3 1 L2 h h 2 I . 1+ 2 12 k A 5 k A2 (17)

trip time (ms)

10

1 1 I/Io 10

Figure 9. Trip time versus over-current ratio: experimental values (symbols) and model (solid line).

Without the second term, this is the same expression as is obtained if temperature dependence of resistivity is neglected. In such a case, since the expansion varies linearly with and as I 2 , the fractional temperature sensitivity of trip current resulting from resistivity variation will be half the fractional sensitivity of itself. Taking 0.6% K1 for the latter, we would obtain 0.3% K1 for the former, about double that observed. However, the additional term in (17) brings us closer to the measured result: using parameter values given above, we obtain from (17) that (dI/I )/dT = 0.39(d/ )/dT, giving 0.23% K1 for the trip current, reasonably close to the 0.15% K1 measured. The model incorporating temperature dependence of resistivity changes the temperature distribution, and results in a higher peak temperature than is obtained with the simpler model. This is shown in gure 8 for several current values, and other parameter values as given above. It can also be seen that the peak temperatures are not excessive in terms of potential damage to the structure. Note that we have taken no account in the analysis above of the temperature dependence of the thermal expansion coefcient. This dependence is, however, relatively small, being about 0.07% K1 for pure Ni in the temperature range of interest [10].

approximations as follows. As the temperature distribution approaches equilibrium, the rate of change will drop, as will the rate of increase of the actuation force. If the equilibrium value of Fe for a given current is denoted by Feo, then we can approximate the rate of change of this latch release force by: Fe (t) = Feo (1 et/ ). (18) We then need to estimate the time constant . This we can do by approximating the rate of increase of Fe at t = 0, and equating this to Feo/ (the initial slope of (18)). At the initial stages of heating, the temperature rise will be relatively uniform along the hot arm. Then the integral in (1) is given simply by (T(x) TA)Lh. Since the conductive cooling is initially negligible, the rise in excess heat per unit volume can be equated to the electrical power according to dT I 2R (19) = dt ALh where m and cp are the mass density and heat capacity respectively. We can then derive dFe(t)/dt from dT/dt, and thus obtain our approximation to as 1 m cp 2 Lh . = (20) 12 k If the required trip force is FeT, the trip time can be derived from (4) as ln(Feo/(Feo FeT)). Since the equilibrium force is proportional to I 2 , we obtain m cp t = ln I2 I2 . 2 Io (21)

5. Trip speed
The speed of tripping is an important performance gure. Short of a full dynamic analysis, we can obtain useful analytic

Taking values for m, cp and k of 8900 kg m3, 0.45 kJ kg1 K1 and 900 W m1 K1 respectively [8], and Lh = 2 mm, the time constant = 18 ms. For large over-current ratios I/Io, (7) reduces to t = (Io/I )2. S113

E M Yeatman

MEMS Dual Breaker

V1 D1 V2 0V C1 C2 B2 A1 A2 B1 D2 E1 E2

Load Circuit

Breaker Control Circuit

Figure 10. Block diagram of possible implementation.

Trip times were measured as a function of the ratio of applied current I to minimum trip current Io. The results are shown in gure 9, along with the theoretical prediction of (21). The t is close for lower currents, but for over-current ratios above 2, the trip time is longer than predicted and appears to level off. One possible cause is the time taken for the mechanism to displace the small distance of the latch mating surfaces (20 m). However, this time can be shown to be insignicant if inertial factors dominate. The total mass of the actuators is very low (0.3 mg), and the driving force will be a signicant fraction of Feo, since the dynamic coefcient of friction is signicantly less than the static one. Taking the driving force as 100 N gives a time to release the latch of 40 s. A non-inertial sticking effect in the latch may instead be the cause. For comparison, nominal trip times for PTCs of similar current rating are typically in the 100 to 1000 ms range [11]. However, these nominal times are usually rated at 510 times the nominal load current, and the actual trip speeds depend very strongly on the fractional over-current. Fault currents of at least three times the nominal load value are typically required in PTCs to achieve even reasonable trip speed.

dependence, and rapid switching for low over-current ratios, are two of these, as discussed above. The ability to design for a wide range of low current applications, according to the design methodology presented here, is another. Also, the MEMS device presents a very low series capacitance in the off state (open circuit), which may be of value in blocking high frequency signals. Measured open circuit capacitances were found to be 0.8 0.2 pF; as these values are dominated by parasitic capacitances of both terminals to the substrate, they can be reduced further by increasing the substrate isolation. Figure 10 shows a possible implementation of the device. A load circuit requires two line voltages, and these are connected through a dual MEMS circuit breaker chip, on which two devices as shown in gure 1 are nested. The pin labels on this chip correspond to those of gure 1. A control circuit monitors the line voltages at E1 and E2 in order to detect any trip event, and is connected to the load circuit from which additional status information can be obtained. On the basis of this status information the decision is made regarding restoring the circuit, and as appropriate the control circuit can then apply reset currents at A1 and A2.

6. Discussion and conclusions

MEMS circuit breakers have been described on the basis of thermal bimorph actuators. Detailed thermo-mechanical analysis provides a basis for designing variants of the device for different circuit protection applications, showing tradeoffs between parameters such as device size, trip and set currents, and speed. The trip current has a low temperature dependence compared to PTC devices; this temperature dependence is analysed and found to be due mainly to resistivity variation. A model for trip time is presented which gives good correspondence to measurements for low over-current ratios, but trip times are found to drop less quickly than predicted for high over-currents. Factors remaining to be studied in detail include lifetime and reliability issues, such as the ability of the contacts to withstand multiple trip and set cycles at high load currents. MEMS circuit breakers offer a number of potential advantages over current solutions. Low temperature S114

Acknowledgments
The author is grateful to the EPSRC and the Department of Trade and Industry for support of this work, to M Geear for experimental measurements, and to the anonymous referees for useful suggestions.

References
[1] Qui J, Lang J H, Slocum A H and Strumpler R 2003 A high-current electrothermal bistable MEMS relay Proc. 16th IEEE Int. Conf. on Micro Electro Mechanical Systems pp 647 [2] Sun X-Q, Farmer K R and Carr W N 1998 A bistable microrelay based on two-segment multimorph cantilever actuators Proc. IEEE MEMS Workshop pp 1549 [3] Wiener P and Witham J 1997 Meeting USB and IEEE 1394 overcurrent protection requirements using PolySwitch devices Proc. IEEE WESCON/97 (Nov. 1997) vols 46 pp 44275

Design and performance analysis of thermally actuated MEMS circuit breakers

[4] Geear M, Yeatman E M, Holmes A S, Syms R R A and Finlay A P 2004 Microengineered electrically resettable circuit breaker J. Microelectromech. Syst. 13 88794 [5] Guckel H, Klein J, Christenson T, Skrobis K, Laudon M and Lovell E G 1992 Thermo-magnetic metal exure actuators Proc. 5th IEEE Solid-State Sensor and Actuator Workshop pp 735 [6] Hyman D and Mehregany M 1999 Contact physics of gold microcontacts for MEMS switches IEEE Trans. Compon. Packag. Technol. 22 35764

[7] Hasee A S M A, Celis J P and Roos J R 2003 Fretting wear of metallic multilayer lms Thin Solid Films 444 99207 [8] Weast R C (ed) 1985 Handbook of Chemistry and Physics (Boca Raton, FL: CRC Press) [9] Minco Application Aid # 18 2000 Minco Products Inc., Minneapolis, MN, USA [10] Data sheet: Super-pure nickel, Isabellenhuette Heusler GmBh & Co., Dillenburg, Germany [11] Multifuse R PTC Resettable Fuses Solutions Guide, Rev. 3.3 2002 Bourns Inc.

S115