IEEE

Proo

f

IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS 1

Two-Axis MEMS Lens Alignment System forFree-Space Optical Interconnect

1

2

Brian E. Yoxall, Member, IEEE, Robert Walmsley, Huei-Pei Kuo, Shih-Yuan Wang, Fellow, IEEE, Mike Tan,and David A. Horsley, Member, IEEE

3

4

Abstract—We present a two-axis microelectromechanical sys-5tems (MEMS) lens aligner with a 260 μm × 220 μm transla-6tion range that positions a 6.35 mm lens with focal length f =712.1 mm for alignment compensation of free-space optical inter-8connects (FSOIs) between computer servers separated by 50 mm9spacing. Efficient ultrasonic linear piezoelectric motors (PMs) pro-10vide actuation with zero power required to hold the lens alignment.11A four-channel FSOI is demonstrated with 1 × 4 vertical cavity12surface-emitting laser (VCSEL) and photodiode (PD) arrays capa-13ble of 10 Gb/s transmission bandwidth. Demonstrated minimum14step size of 1.68 μm is sufficient for aligning the 20 μm VCSEL15spots onto the 40 μm PD receivers. Force transmission between16PMs and a silicon MEMS flexure stage is accomplished using a171-mm steel ball bearing in a magnetic groove, providing compli-18ance in the nondriven axis. The ball-coupling design has 15 μm19backlash and induces a maximum of 8 μm of cross-axis motion.20

Index Terms—Alignment stage, microelectromechanical systems21(MEMS), optical interconnect, piezoelectric ultrasonic stepper mo-22tor, two-axis actuator.23

I. INTRODUCTION24

BANDWIDTH capacity and high power consumption are25

fundamental limitations of the traditional copper-wire-26

based communication in computer servers. At very high band-27

widths, optical interconnects will require less power to operate28

than copper data paths [1]. Optical interconnects are already29

widely used at the cabinet level in data centers and are promis-30

ing candidates for on-chip interconnects in future microproces-31

sors [2]. Here, we consider the problem of creating free-space32

optical interconnect (FSOI) at the board-to-board level. In com-33

parison to interconnects using on-board waveguides routed to34

an optical backplane, FSOI offers the advantages of high den-35

sity [3] and lower board and backplane cost, but requires active36

multiaxis alignment to achieve efficient optical coupling from37

source to receiver [4]. The causes of source–receiver misalign-38

ment include vibration, thermal shifts, and static misalignment39

between the two server boards. Experimental measurements of40

Manuscript received June 26, 2010; revised September 18, 2010; acceptedSeptember 13, 2010. This work was supported by Hewlett-Packard.

Q1 B. E. Yoxall and D. A. Horsley are with the Department of Mechanicaland Aerospace Engineering, University of California, Davis, CA 95616 USA(e-mail: beyoxall@ucdavis.edu; dahorsley@ucdavis.edu).

R. Walmsley, H. Kuo, S. Y. Wang, and M. Tan are with Hewlett-PackardLaboratory, Palo Alto, CA 94304 USA (e-mail: bob.walmsley@hp.com;hueipei.kuo@hp.com; sywang@hp.com; mike.tan@hp.com).

Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/JSTQE.2010.2089043

a typical server chassis [5] show static misalignment (∼250 μm) 41

is considerably greater than thermal shift (∼20 μm upon pow- 42

ering on the server) or mechanical vibration (<1 μm). As a 43

result, continuous dynamic correction is not required; instead, 44

the ideal alignment system would correct the alignment once 45

(upon installation or replacement of the server), after which the 46

alignment would be retained with zero power consumption. 47

Earlier approaches to alignment systems for board-to-board 48

FSOI have included bulk prisms [6], beam splitters [7], mechan- 49

ical translation stages [8], liquid crystal beam steering [9], and 50

optical microelectromechanical systems (MEMS) devices [10]. 51

Among these, optical MEMS technology is attractive because it 52

allows wafer-scale batch fabrication of low-cost devices. Many 53

existing two-axis MEMS lens scanner designs require constant 54

power to maintain lens position [11], have a limited correction 55

range [12], or have a relatively small lens diameter [13], making 56

them incapable of aligning an array of lasers over the 50-mm 57

distance separating two boards in a modern server chassis. 58

Here, we describe a MEMS-based low-power active align- 59

ment system for FSOI using a piezoelectrically actuated lens 60

to align vertical cavity surface-emitting laser (VCSEL) arrays 61

to receiver photodiodes (PDs). The piezoelectric stepper motor 62

technology used here consumes less than 250 mW in full-power 63

operation and requires zero power to maintain static alignment, 64

similar to recent MEMS stepper motor designs [14], [15]. Unlike 65

many earlier MEMS lens alignment systems, the bulk microma- 66

chined stage described here is suitable for positioning a large 67

6.35-mm diameter lens, allowing the alignment of 1× 4 VCSEL 68

arrays over a 50-mm link distance. 69

II. DESIGN 70

A. Device and Test Setup 71

The lens aligner is a 1.3 cm × 3.3 cm × 3.3 cm assembly of 72

a silicon MEMS flexure stage, a plastic aspheric lens, and two 73

piezoelectric motors (PMs), as shown in Fig. 1. The spacing 74

and alignment of the flexure stage and PMs is provided by 75

an aluminum mounting block. Force is transmitted between 76

the MEMS flexure and PMs by a ball/groove device, which 77

decouples the two translation axes, as described in Section II-D. 78

The board-to-board schematic is shown in Fig. 2. The MEMS 79

aligner translates the first lens of a telecentric lens pair to posi- 80

tion VCSEL beams to the PD array. The position information 81

from the PD array is used for feedback control. The telecen- 82

tric lens configuration reduces distortion and sensitivity to axial 83

misalignment and board-to-board spacing variation [16]–[18]. 84

1077-260X/$26.00 © 2010 IEEE

IEEE

Proo

f

2 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS

Fig. 1. Assembly schematic for complete two-axis lens aligner includingmonolithic mount, PMs and ball/groove couplers, MEMS flexure stage, as-pheric lens, and PC Board.

Fig. 2. Test setup to demonstrate feedback control of the MEMS/FSOI align-ment correction device. The primary components are VCSEL source, MEMSlens aligner, receiver lens, and PD array.

B. Piezoelectric Stepper Motor Characterization85

Ultrasonic PMs are ideal for MEMS actuation due to their86

high driving force, large stroke, small incremental motion, and87

low power requirements [19]. The PMs used here (PI-652,88

Physik Instrumente) are 9 mm × 5.4 mm linear motors with89

3-mm actuation range and minimum step size less than 1 mm.90

The PM has a maximum actuating force of 0.1 N and a frictional91

hold force of 0.2 N when not driven. The PM consists of a mov-92

ing carriage/slider mounted on a resonating piezoelectric stator,93

as shown in Fig. 3. Electrodes on the two halves of the stator94

allow excitation at its resonant frequency (450 kHz), causing the95

carriage to travel toward the right or left, depending on which96

Fig. 3. PM image showing the piezoelectric stator and steel carriage/slider.The stator operates in a resonant mode that incrementally moves the slider aslittle as ∼1 μm.

Fig. 4. (a) Step size as a function of drive pulse duration. (b) LDV measure-ment on oscilloscope of PM vibration during a drive cycle, showing ring uptime ∼0.15 ms. The minimum drive pulse is 0.1 ms, during which time, the PMvibration never reaches full amplitude, but it is sufficient to cause slider motion.

electrode is driven. The duration of the sinusoidal excitation sig- 97

nal is gated using a TTL input pulse signal. As described later, 98

the carriage actuation distance is a function of the duration and 99

amplitude of excitation as well as the load and position on the 100

stator. 101

To characterize the mechanical behavior of the PM, a laser 102

Doppler vibrometer (LDV) was used to measure the time- 103

dependent displacement of the stator edge in response to a 104

0.5-ms pulse input. As shown in Fig. 4(b), the steady-state vi- 105

bration amplitude is reached 150 μs after the start of excitation, 106

corresponding to a mechanical quality factor Q ∼ 100. LDV 107

measurements collected at varying supply voltages showed that 108

the vibration amplitude scaled in proportion to the supply volt- 109

age. Carriage motion is small until the vibration is near the 110

steady-state amplitude, thus the minimum excitation pulse is 111

0.1 ms. Measurements collected at a supply voltage of 3.4 V for 112

0.1, 0.2, and 0.3 ms excitation pulses showed a nearly linear de- 113

pendence of the motion on the pulse duration, with a minimum 114

step size of 1.68 μm, as shown in Fig. 4(a). Similarly, varying 115

the supply voltage from 3.4 to 5 V for a constant 0.1 ms pulse 116

duration resulted in a nearly linearly increase in the minimum 117

step size from 1.68 to 11 μm. 118

As a measure of the repeatability of positioning, Fig. 5 shows 119

a return-to-home (RTH) test in which the PM was driven at 120

the minimum step size for ten steps in each direction. Over 121

IEEE

Proo

f

YOXALL et al.: TWO-AXIS MEMS LENS ALIGNMENT SYSTEM FOR FREE-SPACE OPTICAL INTERCONNECT 3

Fig. 5. Drift in position of PM rotor over 20 identically driven RTH tests (tenminimum size steps in each direction) is larger than the minimum step distance.Measurements taken with unloaded PM (no MEMS flexure).

Fig. 6. (a) Folded flexure design uses two parallel sets of beams of length L andwidth w, shown in deflected position. In this image, L = 2 mm and w = 30 μm.(b) Two folded flexure pairs are used for each axis and the ball/groove couplerstransmit PM force to the MEMS flexure.

the course of 20 RTH tests, the zero position of the PM rotor122

moved over 10 μm. Note that with 0.1 ms drive pulse, the stator123

vibration never reaches full amplitude, so the step size is likely124

to be more sensitive to stiction or other perturbations than a125

longer duration drive.126

C. MEMS Flexure Stage127

The MEMS flexure stage, as shown in Fig. 6(b), is composed128

of an inner frame, which provides y-axis movement, and an129

outer frame, which provides x-axis movement. The 20 mm ×130

20 mm device is fabricated from a 300-μm thick (1 0 0) Si131

wafer using a single-mask bulk micromachining process based132

on deep-reactive-ion etching (DRIE) through the entire wafer133

thickness. A plastic aspheric lens with focal length 12.1 mm,134

clear aperture of 4.4 mm, diameter of 6.35 mm, and mass 50 mg135

is affixed to the inner frame of the stage.136

The flexures are compliant in-plane and stiff out-of-plane. To137

avoid introducing misalignment from lens vibration, the stage138

is designed to have low sensitivity to vibrations (due to cooling139

fans, hard-drives, etc.) up to ∼1 kHz. While in-plane stiffness of140

the stage is determined primarily by coupling with the PMs, out-141

of-plane stiffness is determined by the silicon flexures design.142

A folded flexure design is used, as shown in Fig. 6(a), to avoid143

nonlinear stiffening over a large (> 100 μm) displacement range144

while maintaining a high cross-axis stiffness ratio [20]. The145

stiffness of the folded flexure suspension is given by146

keq =48ESiIz

L3 (1)

Iz =hw3

12(2)

Fig. 7. Resonance measurement of flexure stage before adding lens. Plot isamplitude of device motion measured by LDV versus excitation frequency.Inner axis resonance is at ∼487 Hz, outer axis resonance is at ∼360 Hz.

where ESi = 169 GPa is Young’s modulus for single crystal 147

silicon along the 〈1 1 0〉 crystalline axes, while h, w, and L 148

represent the height, width, and length of the individual flexure 149

beams, respectively. Because a stiffness relationship similar to 150

(1) holds for out-of-plane (z-axis) deflections, the ratio of the 151

out-of-plane to in-plane stiffnesses is approximately given by 152

(h/w)2 . The maximum value of (h/w) for defect-free DRIE 153

fabrication is 10:1, so for h = 300 μm wafer thickness, w = 154

30 μm. 155

Given the maximum rated force for each PM is 0.1 N, the stiff- 156

ness of the flexures should be no more than 400 N/m to allow 157

a full-scale lens displacement of xmax = ±250 μm to compen- 158

sate for static board-to-board misalignment [5]. At the same 159

time, it is desirable to make the flexures as stiff as possible to 160

increase the frequency of the first mechanical resonance. For the 161

prototype devices presented here, flexure length L = 2 mm was 162

selected. Substituting this value into (1) yields keq = 684 N/m, 163

allowing a maximum displacement of xmax = 146 μm assum- 164

ing Fmax = 0.1 N. 165

Experimental measurements of the flexure stiffness were per- 166

formed by mounting the MEMS device on a vibration exciter 167

and observing the motion of the inner and outer frames using 168

an LDV. Swept sine measurements performed over a frequency 169

range of 100 Hz to 1 kHz, as shown in Fig. 7, show the in-plane 170

resonances of the inner (y-axis) and outer (x-axis) frames at 171

487 and 360 Hz, respectively, corresponding to stiffnesses of 172

ky = 254 N/m and kx = 279 N/m. 173

The fabricated flexures were overetched and undercut with 174

flexure top width a = 27.5 μm, and bottom width b = 16.0 μm. 175

Iz for a trapezoidal beam cross section is given by 176

Iz=h(a + b)(a2 + b2)

48. (3)

Substituting (3) in (1) gives kcalc = 279 N/m, in good agree- 177

ment with the measured values. 178

D. MEMS/PM Coupling Device 179

Coupling the PM actuators and MEMS flexure stage requires 180

a mechanism that transmits actuator force in the direction of 181

motion but is compliant in the orthogonal direction. In surface 182

micromachined MEMS devices, this coupling has been achieved 183

IEEE

Proo

f

4 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS

Fig. 8. (a) Ball/groove coupler uses a steel ball, held in place by the magneticgroove, to transmit PM force to MEMS flexure stage in the driving directionwhile permitting unrestrained motion in the orthogonal direction. (b) Image ofball and magnets visible through MEMS stage.

by a pin-in-slot mechanism [15]. Here, the coupling relies on a184

precision steel ball bearing (Øball = 1000 ± 2.5 μm) retained185

in a magnetic groove aligned perpendicular to the driven axis,186

as shown in Fig. 8. The inner and outer frames of the stage have187

a square opening (lside = 1005 ± 5 μm) to accommodate the188

steel balls.189

Motion of the outer frame causes the inner frame ball to roll190

along the magnetic groove (and vice versa). Published data on191

the wear of silicon surfaces used to support ball bearings [21]192

suggest a high degree of reliability for the ball–silicon interface.193

The design is tolerant of translational misalignments between194

the PMs and the flexure stage, since each ball can be freely195

positioned during assembly. Two nonlinearities result from the196

design and are explored in the results section. First, clearance197

between the ball and the flexure opening results in a dead zone.198

Second, rotational misalignments between the PM coupler and199

the flexure stage result in cross-axis coupling.200

III. RESULTS201

A. FSOI Setup202

Experiments characterized the optical and mechanical perfor-203

mance of the FSOI system consisting of a source and receiver204

board separated by 50 mm. The receiver board is a 1 × 4 GaAs205

PIN PD array with a fixed lens (f = 12.1 mm), while the source206

board is a 1 × 4 VCSEL array assembled on a printed circuit207

board beneath the MEMS lens aligner, as shown in Fig. 2. Im-208

ages of the VCSEL and PD arrays, as shown in Fig. 9, show that209

both devices have a 250-μm pitch, while the active area of each210

PD is approximately 40 μm.211

Images of the VCSEL spots in the receiver focal plane were212

collected using a CMOS imager (ISG, LW5–5-1394) in place of213

the PD array. Proper alignment between the two lenses, the VC-214

SEL array, and the camera was ensured by collecting images of215

the VCSELs operating below the lasing threshold. Below thresh-216

old, satellite beams are visible around the main emission lobe;217

these beams are clearly visible when proper focus is achieved,218

as shown in Fig. 10.219

The lasing VCSEL beam profiles at the receiver plane with220

the lens aligner positioned such that the beams are near the221

Fig. 9. (a) Microscope image of four VCSEL array at 0.250 mm spacing,with inset image of single VCSEL device. (b) Microscope image of PD array at0.250 mm spacing.

Fig. 10. VCSEL array operating below threshold current in LED mode arecharacterized by satellite spots corresponding to the physical configuration ofthe device.

Fig. 11. VCSEL beam profile measurements taken on CMOS camera for fourlaser spots. The beam diameters are less than 20 μm FWHM.

center of the source lens are shown in Fig. 11. The full width 222

at 10% of maximum intensity is approximately 20 μm. Taken 223

together with the 40 μm PD diameter, the 3 dB radial alignment 224

tolerance of the system is 60 μm, requiring a minimum step size 225

of 42 μm on each axis. 226

As the lens is displaced by the aligner, the focused spots 227

show distortion due to off-axis aberrations in the lens. Fig. 12 228

is a composite image of the VCSEL array in four positions 229

separated by 250 μm. The spot at the lower right shows the 230

maximum distortion and is located ∼500 μm from the lens cen- 231

ter, representative of the position of the spot with the lens aligner 232

driven to its maximum travel in both axes. To avoid saturating 233

the camera, VCSEL images were collected with an aperture and 234

neutral density filters between the source and receiver. Clipping 235

on aperture drops the intensity of the lower right spot to ∼50% 236

of the intensity at lens center, but the beam width of the main 237

IEEE

Proo

f

YOXALL et al.: TWO-AXIS MEMS LENS ALIGNMENT SYSTEM FOR FREE-SPACE OPTICAL INTERCONNECT 5

Fig. 12. Four-by-Four composite image of VCSEL beam distortion as a func-tion of position relative to the telecentric lens pair centerline. The top inset(zoomed image of the spot in row 2, column 2) is the VCSEL beam closelyaligned with the optical axis; the bottom inset (row 4, column 4) shows thedistortion and loss of intensity for a spot ∼0.5 mm from the lens center due toaperture clipping.

Fig. 13. Eye diagrams taken at 2.5 Gb/s data rate at zero offset optical align-ment and 250 μm offset show negligible performance difference.

lobe is similar to undistorted beam width. Measurements col-238

lected with the PD array (without an aperture) showed the peak239

spot intensity varied by less than 6% over a 2.5-mm translation240

range.241

Eye diagrams at 2.5 Gb/s data transmission rate, as shown in242

Fig. 13, have negligible change in quality for a lens translation243

of 250 μm, or the full capability of the system. The eye diagrams244

are taken without any aperture in the beam path. Four parallel245

2.5 Gb/s channels provide for 10 Gb/s communication.246

B. Lens Aligner Performance247

Lens aligner performance was characterized in the FSOI setup248

with the PMs driven with 0.1 ms pulses to achieve the minimum249

step size. Fig. 14 shows overlaid images of a single VCSEL250

spot translated by lens actuation to the extreme corners of the251

MEMS/PM range. Lens motion is 1:1 with image motion on the252

target due to the telecentric lens design. The observed limits of253

travel, ±110 μm in the x-axis and ±130 μm in the y-axis, are254

consistent with a maximum PM output force of 31 and 33 mN,255

roughly 1/3 the manufacturer value.256

The variation of the step size over the full actuation range was257

measured by first driving the stage to extreme limit of travel,258

and then, stepping the lens to the opposite limit, while tracking259

the centroid of the VCSEL spot following each step command,260

as shown in Fig. 15. The observed behavior is approximated261

Fig. 14. Composite image of a single VCSEL spot showing the limits of travelfor the MEMS lens aligner.

Fig. 15. Stage motion over full range of travel. The dashed line shows thestep size predicted based on the net force on the flexure stage. Step size near thebacklash region is ∼2.2 μm.

with a model in which the step size Δ is proportional to the net 262

force acting on the stage 263

Δ = δ (Fmax − kxx) (4)

where δ is a constant of proportionality and is equal to 48 mm/N 264

for the inner axis. The PMs have nonuniform force charac- 265

teristics across the full range, with less force available when 266

operating near full stroke. The outer axis PM operates near 267

full stroke resulting in a much lower proportionality constant 268

d = 8 mm/N. At the initiation of motion, the forces from the 269

PM and the flexures are additive and the largest step size (12 μm) 270

is observed, whereas the step size diminishes to a value below 271

100 nm at the opposite extreme of travel. It was observed that 272

the position of the PM slider on the stator effects δ. One conse- 273

quence of this fact is that the displacement range is asymmetric, 274

since the slider is not centered on the stator when the flexure 275

stage is at zero deflection. 276

The backlash region of stage motion is observable in Fig. 15 277

when the stage reaches zero deflection. The size of the region is 278

calculated as the product of the slope adjacent to the backlash 279

(flat region) with the number of steps in the backlash region. 280

For the inner axis, the backlash is calculated as 15.4 μm, while 281

IEEE

Proo

f

6 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS

Fig. 16. Centroid motion of single VCSEL incident on CMOS camera forMEMS flexure stage motion in both axes, at full travel. Vertical motion cor-responds to MEMS flexure stage inner axis, horizontal motion corresponds toMEMS flexure stage outer axis.

Fig. 17. Nonlinearity in outer axis motion is due to cross-coupling of inner axisand ball/groove interaction with MEMS structure. (a) Misalignment betweenthe groove and the direction of outer axis motion causes cross-coupling intoinner axis motion. (b) and (c) Defects in the surface of the groove cause highand low energy regions through the ball travel. As the ball moves past a highenergy location, it loses contact with the left edge of the square opening in theMEMS. (d) As the ball moves to a low energy position, it remain stationaryuntil contact with the MEMS opening returns. (e) Ball will move “normally”until another defect causes nonlinear cross-coupling.

the outer axis backlash is similarly calculated to be 14.9 μm.282

Based on known variations in ball diameters and fabrication283

variations across the MEMS wafer, the expected backlash is284

12.5 μm. Measured backlash is somewhat larger than expected285

due to overetch during fabrication.286

Cross-axis coupling was measured by collecting centroid po-287

sition of a single VCSEL beam incident on the CMOS camera288

for sweeps across the lens aligner’s range in both axes. The cen-289

troid sweeps are shown in Fig. 16. Inner axis motion is fairly290

linear and free from cross-axis coupling but outer axis motion 291

has clear nonlinearities and exhibits up to 8 μm of cross-axis 292

motion. 293

The nonlinearity results from the system geometry and mag- 294

net/ball interaction as the inner axis ball/groove coupling travels 295

in its compliant axis. Fig. 17 shows how the cross-axis coupling 296

is introduced to the system. As the outer axis moves, the ball 297

rolls in the inner axis groove. Micrometer scale imperfections 298

in the groove surface impede the motion of the ball, altering its 299

position within the square opening of the MEMS stage. The ef- 300

fect of these defects is fairly repeatable as the outer axis motion 301

is cycled backward and forward, as shown in Fig. 16. 302

IV. CONCLUSION 303

A MEMS lens aligner was described that allows a 6.35-mm 304

diameter, f = 12.1 mm lens to be positioned in two axes over 305

a 260 μm × 220 μm range, a range that allows compensation 306

of static and thermal misalignments between two boards. The 307

low-cost plastic optics used are suitable for a FSOI between 308

two computer servers separated by 50 mm. The PM actuators 309

used require zero input power to maintain the alignment. The 310

largest step size, observed near the limits of travel, was 12 311

μm, approximately four times smaller than required to align the 312

20 μm VCSEL spot within the 40 μm high-speed PDs used 313

here. Measurements of the step size versus displacement sug- 314

gest that submicrometer step size could be reliably achieved 315

by increasing the stiffness of the silicon flexures, at the cost of 316

reducing the range of travel. Neglecting losses due to reflec- 317

tions from the surfaces of the nonantireflection coated lenses, 318

a maximum intensity loss of 6% was observed for a laser spot 319

positioned 500 μm from the center of the lens, representing the 320

worst case for a spot deflected to the maximum limits of travel. 321

The ball-coupling system used here showed repeatable behavior 322

and a relatively small (15 μm) backlash region near the center 323

of travel. Improved finishing of the ball-groove would reduce 324

the 8 μm cross-axis coupling observed here. 325

ACKNOWLEDGMENT 326

The authors would like to thank M.-L. Chan and G. Jaramillo 327

for their assistance with MEMS fabrication for this paper and 328

also grateful to E. Lau and J. Chou with network analysis testing. 329

REFERENCES 330

[1] H. Cho, P. Kapur, and K. C. Saraswat, “Power comparison between high- 331speed electrical and optical interconnects for interchip communication,” 332J. Lightw. Tech., vol. 22, no. 9, pp. 2021–2033, Sep. 2004. 333

[2] A. V. Krishnamoorthy, H. Ron, Z. Xuezhe, H. Schwetman, L. Jon, P. Koka, 334L. GuoLiang, I. Shubin, and J. E. Cunningham, “Computer systems based 335on silicon photonic interconnects,” Proc. IEEE, vol. 97, no. 7, pp. 1337– 3361361, Jul. 2009. 337

[3] L. J. Camp, R. Sharma, and M. R. Feldman, “Guided-wave and free- 338space optical interconnects for parallel-processing systems: A compari- 339son,” Appl. Opt., vol. 33, no. 26, pp. 6168–6180, Sep. 1994. 340

[4] A. G. Kirk, D. V. Plant, M. H. Ayliffe, M. Chateauneuf, and F. Lacroix, 341“Design rules for highly parallel free-space optical interconnects,” IEEE 342J. Sel. Top Quant. Electron., vol. 9, no. 2, pp. 531–547, Mar./Apr. 2003. 343

[5] H. P. Kuo, P. Rosenberg, R. Walmsley, S. Mathai, L. Kiyama, J. Straznicky, 344M. McLaren, M. Tan, and S.-Y. Wang, “Free-space optical links for 345

IEEE

Proo

f

YOXALL et al.: TWO-AXIS MEMS LENS ALIGNMENT SYSTEM FOR FREE-SPACE OPTICAL INTERCONNECT 7

board-to-board interconnects,” Appl. Phys. A, vol. 95, no. 4, pp. 955–346965, Jun. 2009.347

[6] K. Hirabayashi, T. Yamamoto, S. Hino, Y. Kohama, and K. Tateno, “Op-348tical beam direction compensating system for board-to-board free space349optical interconnection in high-capacity ATM switch,” J. Lightw. Tech.,350vol. 15, no. 5, pp. 874–882, May 1997.351

[7] F. Lacroix, E. Bernier, M. H. Ayliffe, F. A. P. Tooley, D. V. Plant, and352A. G. Kirk, “Implementation of a compact, four-stage, scalable optical353interconnect for photonic backplane applications,” Appl. Opt., vol. 41,354no. 8, pp. 1541–1555, Mar. 2002.355

[8] D. V. Plant, M. B. Venditti, E. Laprise, J. Faucher, K. Razavi, M. teauneuf,356A. G. Kirk, and J. S. Ahearn, “256-channel bidirectional optical intercon-357nect using VCSELs and photodiodes on CMOS,” J. Lightw. Tech., vol. 19,358no. 8, pp. 1093–1103, Aug. 2001.359

[9] C. J. Henderson, B. Robertson, D. G. Leyva, T. D. Wilkinson,360D. C. O’Brien, and G. Faulkner, “Control of a free-space adaptive op-361tical interconnect using a liquid-crystal spatial light modulator for beam362steering,” Opt. Eng., vol. 44, no. 7, pp. 075401–075408, 2005.363

[10] A. Tuantranont, V. M. Bright, J. Zhang, W. Zhang, J. A. Neff, and Y. C. Lee,364“Optical beam steering using MEMS-controllable microlens array,” Sens.365Actuator Phys., vol. 91, no. 3, pp. 363–372, Jul. 2001.366

[11] M. C. Wu and W. L. Fang, “Integrated bi-directional focusing and tracking367actuators in a monolithic device for a MEMS optical pick-up head,”368J. Micromech. Microeng., vol. 16, no. 7, pp. 1290–1297, Jul. 2006.369

[12] K. Hedsten, J. Melin, J. Bengtsson, P. Modh, D. Karlen, B. Lofving,370R. Nilsson, H. Rodjegard, K. Persson, P. Enoksson, F. Nikolajeff, and371G. Andersson, “MEMS-based VCSEL beam steering using replicated372polymer diffractive lens,” Sens. Actuator Phys., vol. 142, no. 1, pp. 336–373345, Mar. 2008.374

[13] J. Chou, K. Yu, D. Horsley, B. Yoxall, S. Mathai, M. R. T. Tan, S.-Y. Wang,375and M. C. Wu, “Robust free space board-to-board optical interconnect with376closed loop MEMS tracking,” Appl. Phys. A, vol. 95, no. 4, pp. 973–982,377Jun. 2009.378

[14] J. B. Chou, K. Yu, D. A. Horsley, S. Mathai, B. Yoxall, M. Tan,379S. Y. Wang, and M. C. Wu, “Electrothermally actuated free space board-380to-board optical interconnect with zero power hold,” in Proc. Transducers,381Denver, CO, 2009, pp. 2202–2205.382

[15] H. Toshiyoshi, G. D. Su, J. LaCosse, and M. C. Wu, “A surface micro-383machined optical scanner array using photoresist lenses fabricated by a384thermal reflow process,” J. Lightw. Tech., vol. 21, no. 7, pp. 1700–1708,385Jul. 2003.386

[16] Y. Liu, B. Robertson, G. C. Boisset, M. H. Ayliffe, R. Iyer, and D. V. Plant,387“Design, implementation, and characterization of a hybrid optical inter-388connect for a four-stage free-space optical backplane demonstrator,” Appl.389Opt., vol. 37, no. 14, pp. 2895–2914, May 1998.390

[17] D. T. Neilson and E. Schenfeld, “Plastic modules for free-space optical391interconnects,” Appl. Opt., vol. 37, no. 14, pp. 2944–2952, May 1998.392

[18] S. Sinzinger and J. Jahns, “Integrated micro-optical imaging system with393a high interconnection capacity fabricated in planar optics,” Appl. Opt.,394vol. 36, no. 20, pp. 4729–4735, Jul. 1997.395

[19] B. Watson, J. Friend, and L. Yeo, “Piezoelectric ultrasonic micro/milli-396scale actuators,” Sens. Actuator Phys., vol. 152, no. 2, pp. 219–233, Jun.3972009.398

[20] R. Legtenberg, A. W. Groeneveld, and M. Elwenspoek, “Comb-drive ac-399tuators for large displacements,” J. Micromech. Microeng., vol. 6, no. 3,400pp. 320–329, Sep. 1996.401

[21] M. McCarthy, C. M. Waits, and R. Ghodssi, “Dynamic friction and wear402in a planar-contact encapsulated microball bearing using an integrated403microturbine,” J. Microelectromech. Syst., vol. 18, no. 2, pp. 263–273,404Apr. 2009.405

Brian E. Yoxall (M’10) recieved the B.S. degree in406engineering and the M.E. degree fromHarvey Mudd407College, Claremont, CA, in 2002 and 2003, respec-408tively. He is currently working toward the Ph.D. de-409gree in the Department of Mechanical and Aeronau-410tical Engineering, University of California, Davis.411

From 2003 to 2007, he was a System Engineer at412Lockheed Martin Space Systems Company, where he413was involved in vibration control, system integration,414and system testing.415

416

Robert Walmsley, photograph and biography not available at the time of pub- 417lication. 418

419

Huei-Pei Kuo, photograph and biography not available at the time of publica- 420tion. 421

422

Shih-Yuan Wang (F’xx) received the B.S. degree in 423engineering physics in 1969, and the Ph.D. degree 424in electrical engineering and computer sciences in 4251977, both from the University of California (UC), 426Berkeley. Q2427

He was at the Space Science Laboratory on a 428K-band maser, and at the Electronic Research Lab- 429oratory on nanostructured tunneling devices at ter- 430ahertz frequencies, both at UC. In 1977, he joined 431Hewlett-Packard (HP) Company, where he was en- 432gaged first in a manufacturing division on MESFET, 433

and then, at HP Laboratories, Palo Alto, CA, and was involved in vertical cavity 434surface-emitting lasers, photodetectors, modulators, GaN light-emitting diodes, 435nanophotonics/electronics, and nanowire devices with a short break at a startup. 436

Dr. Wang is a Fellow of the Optical Society of America. 437438

Mike Tan, photograph and biography not available at the time of publication. 439440

David A. Horsley (M’97) received the B.S., M.S., 441and Ph.D. degrees in mechanical engineering from 442the University of California (UC), Berkeley, in 1992, 4431994, and 1998, respectively. 444

He is currently an Associate Professor in the De- 445partment of Mechanical and Aerospace Engineering, 446UC, Davis, where he has been a Co-Director of the 447Berkeley Sensor and Actuator Center since 2005. He 448also held research and development positions at Di- 449con Fiberoptics, Hewlett-Packard Laboratories, and 450Onix Microsystems. His research interests include 451

microfabricated sensors and actuators with applications in optical microelec- 452tromechanical system, communication, displays, and biological sensors. 453

Prof. Horsley is a recipient of the National Science Foundation CAREER 454Award and the UC Davis College of Engineering’s Outstanding Junior Faculty 455Award. 456

457

IEEE

Proo

f

QUERIES458

Q1: Author: Please provide the Grant/project no. provided by Hewlett-Packard .459

Q2. Author: Please provide the year in which Shih-Yuan Wang became “Fellow member” of the IEEE.460

IEEE

Proo

f

IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS 1

Two-Axis MEMS Lens Alignment System forFree-Space Optical Interconnect

1

2

Brian E. Yoxall, Member, IEEE, Robert Walmsley, Huei-Pei Kuo, Shih-Yuan Wang, Fellow, IEEE, Mike Tan,and David A. Horsley, Member, IEEE

3

4

Abstract—We present a two-axis microelectromechanical sys-5tems (MEMS) lens aligner with a 260 μm × 220 μm transla-6tion range that positions a 6.35 mm lens with focal length f =712.1 mm for alignment compensation of free-space optical inter-8connects (FSOIs) between computer servers separated by 50 mm9spacing. Efficient ultrasonic linear piezoelectric motors (PMs) pro-10vide actuation with zero power required to hold the lens alignment.11A four-channel FSOI is demonstrated with 1 × 4 vertical cavity12surface-emitting laser (VCSEL) and photodiode (PD) arrays capa-13ble of 10 Gb/s transmission bandwidth. Demonstrated minimum14step size of 1.68 μm is sufficient for aligning the 20 μm VCSEL15spots onto the 40 μm PD receivers. Force transmission between16PMs and a silicon MEMS flexure stage is accomplished using a171-mm steel ball bearing in a magnetic groove, providing compli-18ance in the nondriven axis. The ball-coupling design has 15 μm19backlash and induces a maximum of 8 μm of cross-axis motion.20

Index Terms—Alignment stage, microelectromechanical systems21(MEMS), optical interconnect, piezoelectric ultrasonic stepper mo-22tor, two-axis actuator.23

I. INTRODUCTION24

BANDWIDTH capacity and high power consumption are25

fundamental limitations of the traditional copper-wire-26

based communication in computer servers. At very high band-27

widths, optical interconnects will require less power to operate28

than copper data paths [1]. Optical interconnects are already29

widely used at the cabinet level in data centers and are promis-30

ing candidates for on-chip interconnects in future microproces-31

sors [2]. Here, we consider the problem of creating free-space32

optical interconnect (FSOI) at the board-to-board level. In com-33

parison to interconnects using on-board waveguides routed to34

an optical backplane, FSOI offers the advantages of high den-35

sity [3] and lower board and backplane cost, but requires active36

multiaxis alignment to achieve efficient optical coupling from37

source to receiver [4]. The causes of source–receiver misalign-38

ment include vibration, thermal shifts, and static misalignment39

between the two server boards. Experimental measurements of40

Manuscript received June 26, 2010; revised September 18, 2010; acceptedSeptember 13, 2010. This work was supported by Hewlett-Packard.

Q1 B. E. Yoxall and D. A. Horsley are with the Department of Mechanicaland Aerospace Engineering, University of California, Davis, CA 95616 USA(e-mail: beyoxall@ucdavis.edu; dahorsley@ucdavis.edu).

R. Walmsley, H. Kuo, S. Y. Wang, and M. Tan are with Hewlett-PackardLaboratory, Palo Alto, CA 94304 USA (e-mail: bob.walmsley@hp.com;hueipei.kuo@hp.com; sywang@hp.com; mike.tan@hp.com).

Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/JSTQE.2010.2089043

a typical server chassis [5] show static misalignment (∼250 μm) 41

is considerably greater than thermal shift (∼20 μm upon pow- 42

ering on the server) or mechanical vibration (<1 μm). As a 43

result, continuous dynamic correction is not required; instead, 44

the ideal alignment system would correct the alignment once 45

(upon installation or replacement of the server), after which the 46

alignment would be retained with zero power consumption. 47

Earlier approaches to alignment systems for board-to-board 48

FSOI have included bulk prisms [6], beam splitters [7], mechan- 49

ical translation stages [8], liquid crystal beam steering [9], and 50

optical microelectromechanical systems (MEMS) devices [10]. 51

Among these, optical MEMS technology is attractive because it 52

allows wafer-scale batch fabrication of low-cost devices. Many 53

existing two-axis MEMS lens scanner designs require constant 54

power to maintain lens position [11], have a limited correction 55

range [12], or have a relatively small lens diameter [13], making 56

them incapable of aligning an array of lasers over the 50-mm 57

distance separating two boards in a modern server chassis. 58

Here, we describe a MEMS-based low-power active align- 59

ment system for FSOI using a piezoelectrically actuated lens 60

to align vertical cavity surface-emitting laser (VCSEL) arrays 61

to receiver photodiodes (PDs). The piezoelectric stepper motor 62

technology used here consumes less than 250 mW in full-power 63

operation and requires zero power to maintain static alignment, 64

similar to recent MEMS stepper motor designs [14], [15]. Unlike 65

many earlier MEMS lens alignment systems, the bulk microma- 66

chined stage described here is suitable for positioning a large 67

6.35-mm diameter lens, allowing the alignment of 1× 4 VCSEL 68

arrays over a 50-mm link distance. 69

II. DESIGN 70

A. Device and Test Setup 71

The lens aligner is a 1.3 cm × 3.3 cm × 3.3 cm assembly of 72

a silicon MEMS flexure stage, a plastic aspheric lens, and two 73

piezoelectric motors (PMs), as shown in Fig. 1. The spacing 74

and alignment of the flexure stage and PMs is provided by 75

an aluminum mounting block. Force is transmitted between 76

the MEMS flexure and PMs by a ball/groove device, which 77

decouples the two translation axes, as described in Section II-D. 78

The board-to-board schematic is shown in Fig. 2. The MEMS 79

aligner translates the first lens of a telecentric lens pair to posi- 80

tion VCSEL beams to the PD array. The position information 81

from the PD array is used for feedback control. The telecen- 82

tric lens configuration reduces distortion and sensitivity to axial 83

misalignment and board-to-board spacing variation [16]–[18]. 84

1077-260X/$26.00 © 2010 IEEE

IEEE

Proo

f

2 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS

Fig. 1. Assembly schematic for complete two-axis lens aligner includingmonolithic mount, PMs and ball/groove couplers, MEMS flexure stage, as-pheric lens, and PC Board.

Fig. 2. Test setup to demonstrate feedback control of the MEMS/FSOI align-ment correction device. The primary components are VCSEL source, MEMSlens aligner, receiver lens, and PD array.

B. Piezoelectric Stepper Motor Characterization85

Ultrasonic PMs are ideal for MEMS actuation due to their86

high driving force, large stroke, small incremental motion, and87

low power requirements [19]. The PMs used here (PI-652,88

Physik Instrumente) are 9 mm × 5.4 mm linear motors with89

3-mm actuation range and minimum step size less than 1 mm.90

The PM has a maximum actuating force of 0.1 N and a frictional91

hold force of 0.2 N when not driven. The PM consists of a mov-92

ing carriage/slider mounted on a resonating piezoelectric stator,93

as shown in Fig. 3. Electrodes on the two halves of the stator94

allow excitation at its resonant frequency (450 kHz), causing the95

carriage to travel toward the right or left, depending on which96

Fig. 3. PM image showing the piezoelectric stator and steel carriage/slider.The stator operates in a resonant mode that incrementally moves the slider aslittle as ∼1 μm.

Fig. 4. (a) Step size as a function of drive pulse duration. (b) LDV measure-ment on oscilloscope of PM vibration during a drive cycle, showing ring uptime ∼0.15 ms. The minimum drive pulse is 0.1 ms, during which time, the PMvibration never reaches full amplitude, but it is sufficient to cause slider motion.

electrode is driven. The duration of the sinusoidal excitation sig- 97

nal is gated using a TTL input pulse signal. As described later, 98

the carriage actuation distance is a function of the duration and 99

amplitude of excitation as well as the load and position on the 100

stator. 101

To characterize the mechanical behavior of the PM, a laser 102

Doppler vibrometer (LDV) was used to measure the time- 103

dependent displacement of the stator edge in response to a 104

0.5-ms pulse input. As shown in Fig. 4(b), the steady-state vi- 105

bration amplitude is reached 150 μs after the start of excitation, 106

corresponding to a mechanical quality factor Q ∼ 100. LDV 107

measurements collected at varying supply voltages showed that 108

the vibration amplitude scaled in proportion to the supply volt- 109

age. Carriage motion is small until the vibration is near the 110

steady-state amplitude, thus the minimum excitation pulse is 111

0.1 ms. Measurements collected at a supply voltage of 3.4 V for 112

0.1, 0.2, and 0.3 ms excitation pulses showed a nearly linear de- 113

pendence of the motion on the pulse duration, with a minimum 114

step size of 1.68 μm, as shown in Fig. 4(a). Similarly, varying 115

the supply voltage from 3.4 to 5 V for a constant 0.1 ms pulse 116

duration resulted in a nearly linearly increase in the minimum 117

step size from 1.68 to 11 μm. 118

As a measure of the repeatability of positioning, Fig. 5 shows 119

a return-to-home (RTH) test in which the PM was driven at 120

the minimum step size for ten steps in each direction. Over 121

IEEE

Proo

f

YOXALL et al.: TWO-AXIS MEMS LENS ALIGNMENT SYSTEM FOR FREE-SPACE OPTICAL INTERCONNECT 3

Fig. 5. Drift in position of PM rotor over 20 identically driven RTH tests (tenminimum size steps in each direction) is larger than the minimum step distance.Measurements taken with unloaded PM (no MEMS flexure).

Fig. 6. (a) Folded flexure design uses two parallel sets of beams of length L andwidth w, shown in deflected position. In this image, L = 2 mm and w = 30 μm.(b) Two folded flexure pairs are used for each axis and the ball/groove couplerstransmit PM force to the MEMS flexure.

the course of 20 RTH tests, the zero position of the PM rotor122

moved over 10 μm. Note that with 0.1 ms drive pulse, the stator123

vibration never reaches full amplitude, so the step size is likely124

to be more sensitive to stiction or other perturbations than a125

longer duration drive.126

C. MEMS Flexure Stage127

The MEMS flexure stage, as shown in Fig. 6(b), is composed128

of an inner frame, which provides y-axis movement, and an129

outer frame, which provides x-axis movement. The 20 mm ×130

20 mm device is fabricated from a 300-μm thick (1 0 0) Si131

wafer using a single-mask bulk micromachining process based132

on deep-reactive-ion etching (DRIE) through the entire wafer133

thickness. A plastic aspheric lens with focal length 12.1 mm,134

clear aperture of 4.4 mm, diameter of 6.35 mm, and mass 50 mg135

is affixed to the inner frame of the stage.136

The flexures are compliant in-plane and stiff out-of-plane. To137

avoid introducing misalignment from lens vibration, the stage138

is designed to have low sensitivity to vibrations (due to cooling139

fans, hard-drives, etc.) up to ∼1 kHz. While in-plane stiffness of140

the stage is determined primarily by coupling with the PMs, out-141

of-plane stiffness is determined by the silicon flexures design.142

A folded flexure design is used, as shown in Fig. 6(a), to avoid143

nonlinear stiffening over a large (> 100 μm) displacement range144

while maintaining a high cross-axis stiffness ratio [20]. The145

stiffness of the folded flexure suspension is given by146

keq =48ESiIz

L3 (1)

Iz =hw3

12(2)

Fig. 7. Resonance measurement of flexure stage before adding lens. Plot isamplitude of device motion measured by LDV versus excitation frequency.Inner axis resonance is at ∼487 Hz, outer axis resonance is at ∼360 Hz.

where ESi = 169 GPa is Young’s modulus for single crystal 147

silicon along the 〈1 1 0〉 crystalline axes, while h, w, and L 148

represent the height, width, and length of the individual flexure 149

beams, respectively. Because a stiffness relationship similar to 150

(1) holds for out-of-plane (z-axis) deflections, the ratio of the 151

out-of-plane to in-plane stiffnesses is approximately given by 152

(h/w)2 . The maximum value of (h/w) for defect-free DRIE 153

fabrication is 10:1, so for h = 300 μm wafer thickness, w = 154

30 μm. 155

Given the maximum rated force for each PM is 0.1 N, the stiff- 156

ness of the flexures should be no more than 400 N/m to allow 157

a full-scale lens displacement of xmax = ±250 μm to compen- 158

sate for static board-to-board misalignment [5]. At the same 159

time, it is desirable to make the flexures as stiff as possible to 160

increase the frequency of the first mechanical resonance. For the 161

prototype devices presented here, flexure length L = 2 mm was 162

selected. Substituting this value into (1) yields keq = 684 N/m, 163

allowing a maximum displacement of xmax = 146 μm assum- 164

ing Fmax = 0.1 N. 165

Experimental measurements of the flexure stiffness were per- 166

formed by mounting the MEMS device on a vibration exciter 167

and observing the motion of the inner and outer frames using 168

an LDV. Swept sine measurements performed over a frequency 169

range of 100 Hz to 1 kHz, as shown in Fig. 7, show the in-plane 170

resonances of the inner (y-axis) and outer (x-axis) frames at 171

487 and 360 Hz, respectively, corresponding to stiffnesses of 172

ky = 254 N/m and kx = 279 N/m. 173

The fabricated flexures were overetched and undercut with 174

flexure top width a = 27.5 μm, and bottom width b = 16.0 μm. 175

Iz for a trapezoidal beam cross section is given by 176

Iz=h(a + b)(a2 + b2)

48. (3)

Substituting (3) in (1) gives kcalc = 279 N/m, in good agree- 177

ment with the measured values. 178

D. MEMS/PM Coupling Device 179

Coupling the PM actuators and MEMS flexure stage requires 180

a mechanism that transmits actuator force in the direction of 181

motion but is compliant in the orthogonal direction. In surface 182

micromachined MEMS devices, this coupling has been achieved 183

IEEE

Proo

f

4 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS

Fig. 8. (a) Ball/groove coupler uses a steel ball, held in place by the magneticgroove, to transmit PM force to MEMS flexure stage in the driving directionwhile permitting unrestrained motion in the orthogonal direction. (b) Image ofball and magnets visible through MEMS stage.

by a pin-in-slot mechanism [15]. Here, the coupling relies on a184

precision steel ball bearing (Øball = 1000 ± 2.5 μm) retained185

in a magnetic groove aligned perpendicular to the driven axis,186

as shown in Fig. 8. The inner and outer frames of the stage have187

a square opening (lside = 1005 ± 5 μm) to accommodate the188

steel balls.189

Motion of the outer frame causes the inner frame ball to roll190

along the magnetic groove (and vice versa). Published data on191

the wear of silicon surfaces used to support ball bearings [21]192

suggest a high degree of reliability for the ball–silicon interface.193

The design is tolerant of translational misalignments between194

the PMs and the flexure stage, since each ball can be freely195

positioned during assembly. Two nonlinearities result from the196

design and are explored in the results section. First, clearance197

between the ball and the flexure opening results in a dead zone.198

Second, rotational misalignments between the PM coupler and199

the flexure stage result in cross-axis coupling.200

III. RESULTS201

A. FSOI Setup202

Experiments characterized the optical and mechanical perfor-203

mance of the FSOI system consisting of a source and receiver204

board separated by 50 mm. The receiver board is a 1 × 4 GaAs205

PIN PD array with a fixed lens (f = 12.1 mm), while the source206

board is a 1 × 4 VCSEL array assembled on a printed circuit207

board beneath the MEMS lens aligner, as shown in Fig. 2. Im-208

ages of the VCSEL and PD arrays, as shown in Fig. 9, show that209

both devices have a 250-μm pitch, while the active area of each210

PD is approximately 40 μm.211

Images of the VCSEL spots in the receiver focal plane were212

collected using a CMOS imager (ISG, LW5–5-1394) in place of213

the PD array. Proper alignment between the two lenses, the VC-214

SEL array, and the camera was ensured by collecting images of215

the VCSELs operating below the lasing threshold. Below thresh-216

old, satellite beams are visible around the main emission lobe;217

these beams are clearly visible when proper focus is achieved,218

as shown in Fig. 10.219

The lasing VCSEL beam profiles at the receiver plane with220

the lens aligner positioned such that the beams are near the221

Fig. 9. (a) Microscope image of four VCSEL array at 0.250 mm spacing,with inset image of single VCSEL device. (b) Microscope image of PD array at0.250 mm spacing.

Fig. 10. VCSEL array operating below threshold current in LED mode arecharacterized by satellite spots corresponding to the physical configuration ofthe device.

Fig. 11. VCSEL beam profile measurements taken on CMOS camera for fourlaser spots. The beam diameters are less than 20 μm FWHM.

center of the source lens are shown in Fig. 11. The full width 222

at 10% of maximum intensity is approximately 20 μm. Taken 223

together with the 40 μm PD diameter, the 3 dB radial alignment 224

tolerance of the system is 60 μm, requiring a minimum step size 225

of 42 μm on each axis. 226

As the lens is displaced by the aligner, the focused spots 227

show distortion due to off-axis aberrations in the lens. Fig. 12 228

is a composite image of the VCSEL array in four positions 229

separated by 250 μm. The spot at the lower right shows the 230

maximum distortion and is located ∼500 μm from the lens cen- 231

ter, representative of the position of the spot with the lens aligner 232

driven to its maximum travel in both axes. To avoid saturating 233

the camera, VCSEL images were collected with an aperture and 234

neutral density filters between the source and receiver. Clipping 235

on aperture drops the intensity of the lower right spot to ∼50% 236

of the intensity at lens center, but the beam width of the main 237

IEEE

Proo

f

YOXALL et al.: TWO-AXIS MEMS LENS ALIGNMENT SYSTEM FOR FREE-SPACE OPTICAL INTERCONNECT 5

Fig. 12. Four-by-Four composite image of VCSEL beam distortion as a func-tion of position relative to the telecentric lens pair centerline. The top inset(zoomed image of the spot in row 2, column 2) is the VCSEL beam closelyaligned with the optical axis; the bottom inset (row 4, column 4) shows thedistortion and loss of intensity for a spot ∼0.5 mm from the lens center due toaperture clipping.

Fig. 13. Eye diagrams taken at 2.5 Gb/s data rate at zero offset optical align-ment and 250 μm offset show negligible performance difference.

lobe is similar to undistorted beam width. Measurements col-238

lected with the PD array (without an aperture) showed the peak239

spot intensity varied by less than 6% over a 2.5-mm translation240

range.241

Eye diagrams at 2.5 Gb/s data transmission rate, as shown in242

Fig. 13, have negligible change in quality for a lens translation243

of 250 μm, or the full capability of the system. The eye diagrams244

are taken without any aperture in the beam path. Four parallel245

2.5 Gb/s channels provide for 10 Gb/s communication.246

B. Lens Aligner Performance247

Lens aligner performance was characterized in the FSOI setup248

with the PMs driven with 0.1 ms pulses to achieve the minimum249

step size. Fig. 14 shows overlaid images of a single VCSEL250

spot translated by lens actuation to the extreme corners of the251

MEMS/PM range. Lens motion is 1:1 with image motion on the252

target due to the telecentric lens design. The observed limits of253

travel, ±110 μm in the x-axis and ±130 μm in the y-axis, are254

consistent with a maximum PM output force of 31 and 33 mN,255

roughly 1/3 the manufacturer value.256

The variation of the step size over the full actuation range was257

measured by first driving the stage to extreme limit of travel,258

and then, stepping the lens to the opposite limit, while tracking259

the centroid of the VCSEL spot following each step command,260

as shown in Fig. 15. The observed behavior is approximated261

Fig. 14. Composite image of a single VCSEL spot showing the limits of travelfor the MEMS lens aligner.

Fig. 15. Stage motion over full range of travel. The dashed line shows thestep size predicted based on the net force on the flexure stage. Step size near thebacklash region is ∼2.2 μm.

with a model in which the step size Δ is proportional to the net 262

force acting on the stage 263

Δ = δ (Fmax − kxx) (4)

where δ is a constant of proportionality and is equal to 48 mm/N 264

for the inner axis. The PMs have nonuniform force charac- 265

teristics across the full range, with less force available when 266

operating near full stroke. The outer axis PM operates near 267

full stroke resulting in a much lower proportionality constant 268

d = 8 mm/N. At the initiation of motion, the forces from the 269

PM and the flexures are additive and the largest step size (12 μm) 270

is observed, whereas the step size diminishes to a value below 271

100 nm at the opposite extreme of travel. It was observed that 272

the position of the PM slider on the stator effects δ. One conse- 273

quence of this fact is that the displacement range is asymmetric, 274

since the slider is not centered on the stator when the flexure 275

stage is at zero deflection. 276

The backlash region of stage motion is observable in Fig. 15 277

when the stage reaches zero deflection. The size of the region is 278

calculated as the product of the slope adjacent to the backlash 279

(flat region) with the number of steps in the backlash region. 280

For the inner axis, the backlash is calculated as 15.4 μm, while 281

IEEE

Proo

f

6 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS

Fig. 16. Centroid motion of single VCSEL incident on CMOS camera forMEMS flexure stage motion in both axes, at full travel. Vertical motion cor-responds to MEMS flexure stage inner axis, horizontal motion corresponds toMEMS flexure stage outer axis.

Fig. 17. Nonlinearity in outer axis motion is due to cross-coupling of inner axisand ball/groove interaction with MEMS structure. (a) Misalignment betweenthe groove and the direction of outer axis motion causes cross-coupling intoinner axis motion. (b) and (c) Defects in the surface of the groove cause highand low energy regions through the ball travel. As the ball moves past a highenergy location, it loses contact with the left edge of the square opening in theMEMS. (d) As the ball moves to a low energy position, it remain stationaryuntil contact with the MEMS opening returns. (e) Ball will move “normally”until another defect causes nonlinear cross-coupling.

the outer axis backlash is similarly calculated to be 14.9 μm.282

Based on known variations in ball diameters and fabrication283

variations across the MEMS wafer, the expected backlash is284

12.5 μm. Measured backlash is somewhat larger than expected285

due to overetch during fabrication.286

Cross-axis coupling was measured by collecting centroid po-287

sition of a single VCSEL beam incident on the CMOS camera288

for sweeps across the lens aligner’s range in both axes. The cen-289

troid sweeps are shown in Fig. 16. Inner axis motion is fairly290

linear and free from cross-axis coupling but outer axis motion 291

has clear nonlinearities and exhibits up to 8 μm of cross-axis 292

motion. 293

The nonlinearity results from the system geometry and mag- 294

net/ball interaction as the inner axis ball/groove coupling travels 295

in its compliant axis. Fig. 17 shows how the cross-axis coupling 296

is introduced to the system. As the outer axis moves, the ball 297

rolls in the inner axis groove. Micrometer scale imperfections 298

in the groove surface impede the motion of the ball, altering its 299

position within the square opening of the MEMS stage. The ef- 300

fect of these defects is fairly repeatable as the outer axis motion 301

is cycled backward and forward, as shown in Fig. 16. 302

IV. CONCLUSION 303

A MEMS lens aligner was described that allows a 6.35-mm 304

diameter, f = 12.1 mm lens to be positioned in two axes over 305

a 260 μm × 220 μm range, a range that allows compensation 306

of static and thermal misalignments between two boards. The 307

low-cost plastic optics used are suitable for a FSOI between 308

two computer servers separated by 50 mm. The PM actuators 309

used require zero input power to maintain the alignment. The 310

largest step size, observed near the limits of travel, was 12 311

μm, approximately four times smaller than required to align the 312

20 μm VCSEL spot within the 40 μm high-speed PDs used 313

here. Measurements of the step size versus displacement sug- 314

gest that submicrometer step size could be reliably achieved 315

by increasing the stiffness of the silicon flexures, at the cost of 316

reducing the range of travel. Neglecting losses due to reflec- 317

tions from the surfaces of the nonantireflection coated lenses, 318

a maximum intensity loss of 6% was observed for a laser spot 319

positioned 500 μm from the center of the lens, representing the 320

worst case for a spot deflected to the maximum limits of travel. 321

The ball-coupling system used here showed repeatable behavior 322

and a relatively small (15 μm) backlash region near the center 323

of travel. Improved finishing of the ball-groove would reduce 324

the 8 μm cross-axis coupling observed here. 325

ACKNOWLEDGMENT 326

The authors would like to thank M.-L. Chan and G. Jaramillo 327

for their assistance with MEMS fabrication for this paper and 328

also grateful to E. Lau and J. Chou with network analysis testing. 329

REFERENCES 330

[1] H. Cho, P. Kapur, and K. C. Saraswat, “Power comparison between high- 331speed electrical and optical interconnects for interchip communication,” 332J. Lightw. Tech., vol. 22, no. 9, pp. 2021–2033, Sep. 2004. 333

[2] A. V. Krishnamoorthy, H. Ron, Z. Xuezhe, H. Schwetman, L. Jon, P. Koka, 334L. GuoLiang, I. Shubin, and J. E. Cunningham, “Computer systems based 335on silicon photonic interconnects,” Proc. IEEE, vol. 97, no. 7, pp. 1337– 3361361, Jul. 2009. 337

[3] L. J. Camp, R. Sharma, and M. R. Feldman, “Guided-wave and free- 338space optical interconnects for parallel-processing systems: A compari- 339son,” Appl. Opt., vol. 33, no. 26, pp. 6168–6180, Sep. 1994. 340

[4] A. G. Kirk, D. V. Plant, M. H. Ayliffe, M. Chateauneuf, and F. Lacroix, 341“Design rules for highly parallel free-space optical interconnects,” IEEE 342J. Sel. Top Quant. Electron., vol. 9, no. 2, pp. 531–547, Mar./Apr. 2003. 343

[5] H. P. Kuo, P. Rosenberg, R. Walmsley, S. Mathai, L. Kiyama, J. Straznicky, 344M. McLaren, M. Tan, and S.-Y. Wang, “Free-space optical links for 345

IEEE

Proo

f

YOXALL et al.: TWO-AXIS MEMS LENS ALIGNMENT SYSTEM FOR FREE-SPACE OPTICAL INTERCONNECT 7

board-to-board interconnects,” Appl. Phys. A, vol. 95, no. 4, pp. 955–346965, Jun. 2009.347

[6] K. Hirabayashi, T. Yamamoto, S. Hino, Y. Kohama, and K. Tateno, “Op-348tical beam direction compensating system for board-to-board free space349optical interconnection in high-capacity ATM switch,” J. Lightw. Tech.,350vol. 15, no. 5, pp. 874–882, May 1997.351

[7] F. Lacroix, E. Bernier, M. H. Ayliffe, F. A. P. Tooley, D. V. Plant, and352A. G. Kirk, “Implementation of a compact, four-stage, scalable optical353interconnect for photonic backplane applications,” Appl. Opt., vol. 41,354no. 8, pp. 1541–1555, Mar. 2002.355

[8] D. V. Plant, M. B. Venditti, E. Laprise, J. Faucher, K. Razavi, M. teauneuf,356A. G. Kirk, and J. S. Ahearn, “256-channel bidirectional optical intercon-357nect using VCSELs and photodiodes on CMOS,” J. Lightw. Tech., vol. 19,358no. 8, pp. 1093–1103, Aug. 2001.359

[9] C. J. Henderson, B. Robertson, D. G. Leyva, T. D. Wilkinson,360D. C. O’Brien, and G. Faulkner, “Control of a free-space adaptive op-361tical interconnect using a liquid-crystal spatial light modulator for beam362steering,” Opt. Eng., vol. 44, no. 7, pp. 075401–075408, 2005.363

[10] A. Tuantranont, V. M. Bright, J. Zhang, W. Zhang, J. A. Neff, and Y. C. Lee,364“Optical beam steering using MEMS-controllable microlens array,” Sens.365Actuator Phys., vol. 91, no. 3, pp. 363–372, Jul. 2001.366

[11] M. C. Wu and W. L. Fang, “Integrated bi-directional focusing and tracking367actuators in a monolithic device for a MEMS optical pick-up head,”368J. Micromech. Microeng., vol. 16, no. 7, pp. 1290–1297, Jul. 2006.369

[12] K. Hedsten, J. Melin, J. Bengtsson, P. Modh, D. Karlen, B. Lofving,370R. Nilsson, H. Rodjegard, K. Persson, P. Enoksson, F. Nikolajeff, and371G. Andersson, “MEMS-based VCSEL beam steering using replicated372polymer diffractive lens,” Sens. Actuator Phys., vol. 142, no. 1, pp. 336–373345, Mar. 2008.374

[13] J. Chou, K. Yu, D. Horsley, B. Yoxall, S. Mathai, M. R. T. Tan, S.-Y. Wang,375and M. C. Wu, “Robust free space board-to-board optical interconnect with376closed loop MEMS tracking,” Appl. Phys. A, vol. 95, no. 4, pp. 973–982,377Jun. 2009.378

[14] J. B. Chou, K. Yu, D. A. Horsley, S. Mathai, B. Yoxall, M. Tan,379S. Y. Wang, and M. C. Wu, “Electrothermally actuated free space board-380to-board optical interconnect with zero power hold,” in Proc. Transducers,381Denver, CO, 2009, pp. 2202–2205.382

[15] H. Toshiyoshi, G. D. Su, J. LaCosse, and M. C. Wu, “A surface micro-383machined optical scanner array using photoresist lenses fabricated by a384thermal reflow process,” J. Lightw. Tech., vol. 21, no. 7, pp. 1700–1708,385Jul. 2003.386

[16] Y. Liu, B. Robertson, G. C. Boisset, M. H. Ayliffe, R. Iyer, and D. V. Plant,387“Design, implementation, and characterization of a hybrid optical inter-388connect for a four-stage free-space optical backplane demonstrator,” Appl.389Opt., vol. 37, no. 14, pp. 2895–2914, May 1998.390

[17] D. T. Neilson and E. Schenfeld, “Plastic modules for free-space optical391interconnects,” Appl. Opt., vol. 37, no. 14, pp. 2944–2952, May 1998.392

[18] S. Sinzinger and J. Jahns, “Integrated micro-optical imaging system with393a high interconnection capacity fabricated in planar optics,” Appl. Opt.,394vol. 36, no. 20, pp. 4729–4735, Jul. 1997.395

[19] B. Watson, J. Friend, and L. Yeo, “Piezoelectric ultrasonic micro/milli-396scale actuators,” Sens. Actuator Phys., vol. 152, no. 2, pp. 219–233, Jun.3972009.398

[20] R. Legtenberg, A. W. Groeneveld, and M. Elwenspoek, “Comb-drive ac-399tuators for large displacements,” J. Micromech. Microeng., vol. 6, no. 3,400pp. 320–329, Sep. 1996.401

[21] M. McCarthy, C. M. Waits, and R. Ghodssi, “Dynamic friction and wear402in a planar-contact encapsulated microball bearing using an integrated403microturbine,” J. Microelectromech. Syst., vol. 18, no. 2, pp. 263–273,404Apr. 2009.405

Brian E. Yoxall (M’10) recieved the B.S. degree in406engineering and the M.E. degree fromHarvey Mudd407College, Claremont, CA, in 2002 and 2003, respec-408tively. He is currently working toward the Ph.D. de-409gree in the Department of Mechanical and Aeronau-410tical Engineering, University of California, Davis.411

From 2003 to 2007, he was a System Engineer at412Lockheed Martin Space Systems Company, where he413was involved in vibration control, system integration,414and system testing.415

416

Robert Walmsley, photograph and biography not available at the time of pub- 417lication. 418

419

Huei-Pei Kuo, photograph and biography not available at the time of publica- 420tion. 421

422

Shih-Yuan Wang (F’xx) received the B.S. degree in 423engineering physics in 1969, and the Ph.D. degree 424in electrical engineering and computer sciences in 4251977, both from the University of California (UC), 426Berkeley. Q2427

He was at the Space Science Laboratory on a 428K-band maser, and at the Electronic Research Lab- 429oratory on nanostructured tunneling devices at ter- 430ahertz frequencies, both at UC. In 1977, he joined 431Hewlett-Packard (HP) Company, where he was en- 432gaged first in a manufacturing division on MESFET, 433

and then, at HP Laboratories, Palo Alto, CA, and was involved in vertical cavity 434surface-emitting lasers, photodetectors, modulators, GaN light-emitting diodes, 435nanophotonics/electronics, and nanowire devices with a short break at a startup. 436

Dr. Wang is a Fellow of the Optical Society of America. 437438

Mike Tan, photograph and biography not available at the time of publication. 439440

David A. Horsley (M’97) received the B.S., M.S., 441and Ph.D. degrees in mechanical engineering from 442the University of California (UC), Berkeley, in 1992, 4431994, and 1998, respectively. 444

He is currently an Associate Professor in the De- 445partment of Mechanical and Aerospace Engineering, 446UC, Davis, where he has been a Co-Director of the 447Berkeley Sensor and Actuator Center since 2005. He 448also held research and development positions at Di- 449con Fiberoptics, Hewlett-Packard Laboratories, and 450Onix Microsystems. His research interests include 451

microfabricated sensors and actuators with applications in optical microelec- 452tromechanical system, communication, displays, and biological sensors. 453

Prof. Horsley is a recipient of the National Science Foundation CAREER 454Award and the UC Davis College of Engineering’s Outstanding Junior Faculty 455Award. 456

457

IEEE

Proo

f

QUERIES458

Q1: Author: Please provide the Grant/project no. provided by Hewlett-Packard .459

Q2. Author: Please provide the year in which Shih-Yuan Wang became “Fellow member” of the IEEE.460