E L S E V I E R Sensors and Actuators A 62 (1997) 756-759

TU RS A

PHYSICAL

Simulation and fabrication of micromachined cantilever valves

M. Koch *, A.G.R. Evans, A. Brunnschweiler Universi~. of Southampton, Department of Electronics and Computer Science, Highfield, Southampton S0171B J, UK

Abstract

This paper reports the fabrication and simulation of micromachined cantilever valves. The simulation is achieved by coupling the finite- element method (FEM) package ANSYS and the computational fluid dynamics (CFD) package FLOW3D. It is based on iterative runs for the deflection of the cantilever (ANSYS) and the fluid flow in the duct/cantilever area (FLOW3D). A macro programme controls the data flow between the simulators and the convergence of the simulation. With the coupled simulation, flow rates for cantilever valves at given pressures are calculated. These results are compared with micromachined cantilever valves fabricated by fusion bonding of a flap-containing wafer with a duct-containing water. For the release of the cantilever, toluene is used to avoid stiction while drying the chips. Agreement between measured and simulated flow rates is good and gives confidence in the use of this simulator for valve development. © 1997 Published by Elsevier Science S.A.

Ke.~vords: Cantilever valves; Micromachining; Simulations

1. Introduct ion

Cantilever valves (see Fig. 1), i.e., pass ive one-way valves, where a flap over a duct controls the flow, are and will be widely used in micromachining. The applications include micropumps [ 1 ] as well as check valves in chemical analysis systems and future miniatur ized chemical reactors.

For micromaehining, Emmet et al. [2] use fusion bonding of 30 l~m thick silicon cantilevers to a bot tom wafer employ- ing a duct. Zengerle et al. [ i ] stick the canti lever chip and duct chip together at the border of the chips, and metal can- tilever systems were described by Schwes inger et al. [3 ],

Analytical approximations for canti lever valves have been carried out by Park et al. [4] to est imate the flow rate. Fur- thermore, a coupled FEM simulation was already presented by the authors [ 5 ], which allowed flow-rate prediction. How- ever, these simulations involved a lot of t ime-consuming manual file conversion.

In this paper, a computer-control led version is reported. The ad van tage is the complete control b y a macro programme as well as the abil i ty to alter the geometry o f the canti lever valve in the command line. The results of the simulation are compared with measurements made on micromachined can- tilever valves of 500 to 1000 tzm sidelength employing fusion

* Corresponding author. Tel.: +44 1703 593 737. Fax: +44 1703 593 029. E-mail: mk94r@ecs.soton.ac.uk, agre@ecs.soton.ac.uk, ab@ecs. soton.ac.uk

Fig. 1. Principle of the passive cantilever valve.

bonding. The canti lever is formed by the highly boron-doped etch-stop mechanism. After releasing the cantilever in buf- fered HF, toluene was used for the first time to avoid stiction. The results are quite promising, as most cantilever valves were operating after drying.

Measurement and simulation of flow rate are in good agree- ment and show that the simulation is correct. This also sug- gests that the software can be used to develop micromachined cantilever valves to part icular specifications.

2. Fabricat ion

The cantilever valve was fabricated by means of micro- machining. At all stages, batch processing is used. The valve

0924-4247/97/$17.00 © 1997 Published by Elsevier Science S.A. All rights reserved PIIS0924-4247(97)01577-X

M. Koch et al. / Sensors and AcmatorsA 62 (1997) 756-759 757

a)

b)

............................................................................................... i--_ SiO:

,.-...JTi+ ; :7,; ~] I:!' ";1

Alter sawing the individual chips, the intermediate oxide is etched with buffered HF to release the cantilever. Before drying, a sequential rinse in DI water, pure alcohol and finally in toluene is done. Both the operating side of the cantilever and the valve seat have a rather smooth surface. Therefore stiction is a major problem unless the above rinse procedure is executed. The yield after a rinse in toluene is around 80% and gives similar results to freeze-drying or n-hexane tech- niques [7]. Without toluene, no valve operated after drying.

c)

~ D ~ ...... Si sN 4 /lever uct - SiO,_

................. ........... ; . . . . . ; . . . . . . . . . . . . . . , , , ~ . : , ]

d)

Fig. 2. Canti[ever valve fabrication steps.

consists of a cantilever on one wafer sitting on top of an aperture of a through-hole in the second wafer. The fabrica- tion of through-hole and cantilever is done on the same wafer. Using symmetrical placement, the cantilever on one wafer joins the duct of the second wafer.

Because processing is required from the front and back of the wafer surface, an IR aligner which can see through the wafer is used. The process sequence is shown in Fig. 2. Dep- osition and patterning of SiO 2 and Si3N4 is followed by deep B + diffusion at 1150°C (Fig. 2 (a) ) . Boron concentrations above 3 × 10 ~9 cm -3 provide a good etch-stop in KOH bulk etching, and allow the generation of thickness-controlled can- tilevers. The diffusion in this experiment was chosen to yield approximately 4.2 p,m thick cantilevers after KOH etching. The oxide and nitride layer are now replaced by another oxide and nitride layer (see Fig. 2 (b ) ) , which are then patterned for the KOH etch. This etch was performed at 70°C and 10 wt.%. The cantilever is formed at diffused places with a rough surface finish at the etched side and a smooth surface finish at the operating side. The duct is realized where boron dif- fusion was prevented. Fig. 2(c) depicts the etched devices.

If one wafer is now turned and placed above another one, the valve can be completed (Fig. 2 (d) ) . Again, alignment of both is needed to position the cantilever on top of the duct. This is done by sawing two new perpendicular flats along the outer scribelines of the wafer. After oxidation of one wafer and RCA cleaning of both, room-temperature bonding is per- formed on a jig, where the perpendicular flats are aligned. The intermediate oxide (500 rim) is needed to give good bonding quality and to allow the release of the cantilever. The wafers are now annealed in oxygen at 1100°C to yield a permanent bond. Note that cantilever and duct are bonded directly together without the depression used by other authors [6]. This is done to avoid leakage.

3. S i m u l a t i o n a n d m e a s u r e m e n t

3.1. Simulation method

The simulation is a coupled finite-element method/com- putational fluid dynamics (FEM/CFD) simulation, achieved with a macro programme that couples ANSYS (FEM) and FLOW3D (CFD). The macro programme (Fig. 3) converts the output of one simulator to a suitable format for use by the other. It is stm:ted in a command line, where the geomet W of the cantilever and the desired pressure (p~ in Fig. 1) are specified. This includes the possibility of entering an initial deflection of the cantilever, e.g., originating from stress due to non-uniform boron concentrations. For example, an initial deflection of 6 txm at the tip of the 1000 txm long micro- machined flap was measured and entered into the simulation.

INPUT OPERATION OUTPUT

Conm'tand line ] Input: Cantilever S/ze, Pressure

I .............

I eF'e I 1

{ ANSYS: Simulate deflection

,I [,. FLOW3D: Simulate flow

I I Move files to *.old

I FORTRAN: Extract + Convert

~ [ Use cubic spline inTpolalion

I ANSYS: Simulate dellection

Pressure, 1

Logfile ]

Pressure.old Defleet.o d

> Pressure J

> Deflec6on t

No I

Test if difference is < 5%

Test il"diff~renee is < 5% Apply successive . . . . . . overrel&xalion > Det)eetion I

& END: F_.','U'aet flowrate ] ~ Flowiatd ]

Fig. 3, Flow chart for the macro programme.

758 M. Koch et al. / Sensors and Actuators A 62 (1997) 756-759

Two grids are used for the modelling. The ANSYS grid contains a cuboid beam consisting of nearly cubic elements. The nodes were constrained to be fixed at the root of the cantilever. The FLOW3D grid is formed by the duct linked with the area below the cantiIever, as the main pressure loss occurs in this part [5] . The simulation is entirely controlled by the macro programme. An initial condition is specified for the first ANSYS run, to calculate the deflection of the beam by an applied pressure. This pressure is assumed to be the maximum applied pressure in the duct area and zero else- where. Therefore, the macro programme generates a pressure file with the above conditions. ANSYS is then started using the pressure file as an input and calculating the deflection at each node of the cantilever grid. The file containing the deflection array is now translated to the FLOW3D CFD grid using cubic spline interpolation. For the simulation itself, a laminar flow behaviour was used. The output of the FLOW3D simulation is a log file containing pressure and flow velocities for all nodes. Pressure values below the cantilever are then extracted and converted from the FLOW3D logfile into an ANSYS-compatible form. Again cubic spline fitting is used for translation. After this, ANSYS is started with the new pressure file as input, resulting in the new deflection for this pressure condition. Finally, a test is done to evaluate whether the results converge. Therefore, the old and new pressure files as well as the old and new deflection files are compared and the programme finishes if the average and maximum differ- ence is less than a given percentage value. Here, 5 and 10% were taken. In order to accelerate the convergence, non-linear successive over-relaxation was used for the input of the next FLOW3D simulation. The simulation time for a 700 ~xm × 700 Fum cantilever for one iteration was approximately 20 min on a SPARC 1000 computer. On average the simu- lation was finished after three to five iterations.

3.2. Results

Flow-rate measurements of the micromachined valve were taken for pressures between 1 and 8 kPa. Pressure application was achieved by varying the height o f a water column. For the measurements and the simulations, cantilevers of 500 to 1000 p,m sidelength were positioned centrally over an open- ing of 100 ~ m × 100 t~m. DI water (dynamic viscosity /z--- 1.02 × 10 -3 Pa s) was used as the test liquid.

Fig. 4 shows the measured and simulated curves for the different cantilevers. The error bars are determined by vary- ing the thickness and the maximum initial deflection of the cantilever by 0.2 ~m. It is obvious that both simulation and measurement agree very well. This gives confidence that the coupled FEM/CFD simulation is correct over the whole pres- sure range. It was noted that there was no measurable leakage rate when the valve was operated in reverse direction, i .e, when the valve was shut. The minimum detectable flow rate with the apparatus is approximately 1 i.d rain- t.

Simulations for higher pressures were carried out as well. The pressure was varied between 1 and 45 kPa. The results

1200 , '~l '" , . . , r ~,

1000

~ ' ~00

600

400

n-

0 0

200

/ ~ Nu~ttkltioll , l l ~ l ' ~ [(Xl(l~42;tll~ + M~'asurement 10(Xlx Illntl~m

Sil)l tdnfiOll 700 ~: 711q x 4.()Itrll tg Meastlenlent 7g() x "/(~lpm O Simulaticm NI(I X 5IX) g 4.()~.tlT= X M¢~L~tlrl,,nlent 5(1(I X $ ( I ) I t l l l

+ 4 . + +

S

Pressure [Pal Fig. 4. Volume flow of the cantilever valve.

) ~ I 1 ' }

+ + ÷ ' t ' + ~

I 8000

3 5 0 0 , , i . . . . I . . . . t . . . . t . . . . ~ . . . . i . . . . , . . . . i , ' ' ~ . . . .

3000

2500

E 2000

~ 0

,rr-

500

A L a m l n a r s i m u l a t i o n t T tu rbu len t s i m u l a t i o n v

I

I

X

I[

X

cantilever. 700x 7~0 x 42

.~111.1~ I I . . . . I . . . . I , , I . [ I I , , , , L , , , ,

0 sooo ~00 ~o00 2oo00 2sooo aoooo asooo 40000 45000 so000

Pressure (Pa]

Fig, 5. Turbulent and laminar s imulat ion results.

are shown in Fig. 5. The FLOW3D grid for both simulation methods, laminar and turbulent, was identical up to pressures of 20 kPa. In order to achieve convergence of the laminar model for higher pressures, a finer grid was necessary. This increases the calculation time drastically. Since the results of both laminar and turbulent simulations are nearly identical, the faster turbulent model can be used for the simulations at high pressures. An estimation of the Reynolds number for this problem gives a factor of approximately 50 for a maxi- mum cantilever deflection of 50 ~m and a volume flow of 2000 ILl rain- ', being well below the critical Reynolds num- ber (1160) for smooth tubes. The Reynold' s number together with the simulation results both suggest that the inertia terms in the Navier-Stokes equations are playing a minor role except in improving the convergence. The flow can therefore be assumed to be laminar over the whole pressure range.

4. C o n c l u s i o n s

A computer-controlled cantilever-valve simulation system is presented for the first time. Comparisons of measurements and simulations give confidence in the validity of flow-rate results obtained from the simulation tool. This opens the possibility of developing cantilever valves with specified

M. Koch et aL / Sensors and Acmators A 62 (1997) 756-759 759

characteristics, without the time-consuming need for batch processing in microfabrication facilities. Due to computer control with a macro programme, time and costs for research and development can be reduced.

The batch fabrication of a cantilever valve was also described, using toluene to avoid stiction during drying. The highly boron-doped etch-stop mechanism allows thickness- controlled cantilevers, which are essential for valves with specified characteristics, to be made. Fusion bonding without depression for the cantilever guarantees very low leakage rates, while using batch processing.

References

[ l I R, Zengerle, A. Richter and H. Sandmaier, A micro membrane pump with electrostatic actuation, Proc. IEEE Micro Electro Mechanical Systems '92, Travemilnde, Germany, 1992, IEEE, New York, 1992, pp. 19-24.

[2] A. Emmer, M. Jansson, J. Roeraade, U. Lindberg and B. H~Sk, Fabrication and characterization of a silicon microvalve, J. Microcolumn Separation, 4 (1992) i3-15.

[ 3 t N. Schwesinger, T. Frank and H. Wurmus, A modular microfluid system with an integrated micromixer, Proc. Micromechanics Europe 1995, Copenhagen, Dennlark, 1995, pp. 144-147.

[4] S. Park, W.H, Ko and J.M, Prahl, A constant flow-rate microvalve actuator based on silicon and micromachining technology, Tech. Digest, IEEE Solid-State Sensor and Actuator Workshop, Hilton Head Island, SC, USA, 1988, IEEE, New York, pp. 136-139.

[5] M. Koch, A.G.R. Evans and A. Brunnschweiler, Coupled FEM- simulation for the characterization of the fluid flow within a mlcromachined cantilever valve, J. Micromech. Microeng., 6 (1996) 112-i 14.

[6] J. Tiren, L. Tenerez and B. HSk, A batch fabricated non-reverse valve with cantilever beam manufactured by micromachining of silicon, Sensors and Actuators, 18 (1989) 389-396.

[7] P.R. Scheeper, J.A. Voomthuyzen, W. Olthuis and P. Bergveld, Investigation of attractive forces between PECVD silicon nitride microstruetures and an oxidized silicon substrate, Sensors Actuators A, 30 (1992) 231-239.

Biographies

M i c h a e l K o c h was born in ingolstadt, Germany, in 1970. He received the Dipl.-Ing. (FH) degree from the Fachhoch- schute Regensburg, Germany, in 1994. He is currently work- ing towards the Ph.D. degree at Southampton University, UK. His main interest is in the microfluidic area. This includes microvalves, micropumps and microfluidic devices in general.

A l a n E v a n s received the B.Sc. degree in physics from Liverpool University, UK, in 1966 and the D.Phil. degree from Oxford University in 1969. He is at present a reader in the Department of Electronics and Computer Science at Southampton University, UK. His research interests include micromachined silicon sensors and actuators, novel fabrica- tion technology for CMOS devices and conduction mecha- nisms in polysilieon. In recent years he has been responsible for the supervision of industrially funded projects for the development of micromachined pressure sensors using both bulk and polysilicon diaphragms, and more recently has supervised the development of micromachined probes for atomic force microscopy.

A r t h u r B r u n n s c h w e i l e r was born in Manchester, UK, in 1936. He received the B.A. degree from Cambridge Univer- sity in 1956, the M.S. degree from Pennsylvania State Uni- versity in 1960, and the Ph.D. degree from the University of Manchester Institute of Science and Technology in 1966. From 1960 to 1964 he was employed in the Camera Tube Research Department of the English Electric Valve Com- pany, Chelmsford, UK, and since 1966 he has been on the academic staff of Southampton University, UK, where he is currently senior lecturer in the Department of Electronics and Computer Science. His research and teaching interests include most areas of microelectronics, particularly device and circuit design. He has acted as consultant to several indus- trial companies and his present duties include the manage- ment of the Microelectronics Industrial Unit at Southampton University.