close

Вход

Забыли?

вход по аккаунту

?

el%3A19870266

код для вставкиСкачать
Fig. 3 shows far-field patterns of the light emitted from the
mode-filter side of the chip whose deflector-side far-field patterns are shown in Fig. 2. For comparison with Fig. 2, the
naming of directions, left and right, is fixed to the laser chip.
The total current into the deflector and the current into the
mode filter were the same as in Fig. 2. The following two main
features are observed. First, the far-field peak moved as IL and
IR changed. This shows that the far-field patterns are the
superposition of several characteristic modes of the modefiltering waveguide, because the index of the waveguide is
unchanged and the far field of each mode should befixed.In
other words, the function of the mode filter is incomplete. This
strongly suggests that the higher-order mode components
would survive in the light incident on the deflector even after
a round trip through the mode filter. Secondly, the magnitude
of the peak shift was small, and its shifting direction was not
fixed: for all current ratios the main peak appeared within a
small angle region, spreading only by 4°; the peak moved in
the opposite direction near the two extreme cases of (IL, IR) =
(OmA, 80 mA) and (IL, IR) = (80 mA, 0 mA). The fact that the
laser beam emitted from the deflector facet is deflected more
prominently and in a more consistent way than the beam
from the back facet indicates again that the deflection in Fig. 2
is caused mainly in the deflector by the prism effect, and that,
unlike the case of twin-striped lasers, it is not due to the
phase-front-shape deformation occurring with a constant
magnitude all over the cavity length.
right
left
-U
radiation
0
angle, deg
1989/31
Fig. 3 Far-field patterns of laser light emitted from mode-filter facet for
various currents injected into deflector electrode
In conclusion, the deflection of a laser beam using a local
deflector integrated in a laser cavity was demonstrated for the
first time. This beam-deflecting method has, in principle, a
potential advantage of higher controllability compared to the
method using twin-stripe lasers.
The authors are grateful to Y. Noguchi and Y. Mitsuhashi
for discussions.
S. MUKAI
24th February 1987
M. WATANABE
H. ITOH
H. YAJIMA
Electrotechnical Laboratory
1-1-4 Vmezono, Sakura
Niihari, Ibaraki 305, Japan
References
1
SCIFRES, D. R., STREIFER, w., and BURNHAM, R. D.: 'Beam scanning
with twin stripe injection lasers', Appl. Phys. Lett., 1978, 33, pp.
702-704
2
STREIFER, w., BURNHAM, R. D., and SCIFRES, D. R.: 'Symmetrical and
asymmetrical waveguide in a very narrow conducting stripe
lasers', IEEE J. Quantum Electron., 1979, QE-15, pp. 136-141
3
SHORE, K. A., and HARTNETT, P. J.: 'Diffusion and waveguide effects
in twin-striped injection lasers', Opt. & Quantum Electron., 1982,
14, pp.169-176
362
KATZ, J.: 'Electronic beam steering of semiconductor injection
lasers', Appl. Opt., 1983, 22, pp. 313-317
MUKAI, S., WATANABE, M., ITOH, H., YAJIMA, H., HOSOI, Y., a n d
UEKUSA, s.: 'Beam scanning and switching characteristics of twin
striped lasers with a reduced stripe spacing', Opt. & Quantum Electron., 1985,17, pp. 431^34
MUKAI, s., and YAJIMA, H. : 'Analysis of mode behavior in a waveguide with graded index and gain', IEEE J. Quantum Electron.,
1984, QE-20, pp. 728-733
NARROW, HIGH-NA GaAs/GaAIAs OPTICAL
WAVEGUIDES WITH LOSSES BELOW
0-7 1 0 - i d B c m - 1
Indexing terms: Integrated optics, Optical waveguides, Electro-optics
Losses as low as 0-65dBcm" 1 (3/un width) and
(8/mi width) have been measured (X — 115/un) in highconfinement (NA =s 0-45) GaAs/GaAIAs optical waveguides
grown by MOCVD. The Fabry-Perot loss measurement
technique used is deduced to be accurate to i O l d B c m " 1 .
These are the lowest losses reported for guides of high
electro-optic merit in III-V materials.
Introduction: Various workers have reported GaAs/GaAIAs
waveguide propagation losses of l d B c m " 1 or less;1'2
however, these have generally used guides of low electro-optic
merit. We define as a figure of merit for such waveguides the
ratio of phase modulator bandwidth to drive voltage, and find
that narrow, well confined guides offer the highest figures.
Thus, for example, the lowest reported losses to date
(O^dBcm" 1 ) have been for guides 8/mi wide and 6/im deep,1
while other workers using the same width with stronger vertical confinement (~3/mi spot depth) obtained O-TdBcm"1.2
By contrast, the guides measured here produce vertical spot
sizes of around 1-5 /an, the width of interest for high-speed
modulators being about 3/rni. These typically give a phase
modulation slope of 5°V~ 1 mm~ 1 with capacitance of only
310 pFm" 1 . The optical parameters of such waveguides are
well matched to those of semiconductor lasers with consequent application for optical signal processing involving
optoelectronic integration. The high yield of low-loss, centimetre-length guides demonstrated here shows that MOCVD
epitaxy used is suitable for such integration.
Fabry-Perot method: We have previously described a
technique3 which derives the attenuation of a waveguide from
its Fabry-Perot resonances. This requires knowledge only of
the facet reflectivity R (a very stable and predictable
parameter), and is insensitive to the waveguide input coupling
efficiency. It requires a light source which is spectrally stable
and narrow compared with the waveguide resonance spacing.
Here we have used the HeNe 1152nm line. Briefly, the
method is as follows.
After optimisation of the objective input and output coupling, the chip is heated by means of current through a resistor
close to the surface. This is found to sweep the Fabry-Perot
resonances in a controllable manner without affecting the
optical coupling. The guide loss is ultimately derived from the
ratios of transmission maxima to adjacent minima, several of
which are recorded (per guide) and a mean value used. By
repeating the measurement for both sections after recleaving
the chip, the facet reflectivity may be derived.3
Waveguide fabrication: The double heterostructure GaAs/
GaAlAs waveguide layers were grown by MOCVD on 2 in
(50-8 mm) GaAs wafers cut 2° off the [100] towards the [110]
direction for optimal growth. A side-effect of this orientation
is that cleaved facets are off-perpendicular by about 1-4°. The
material was undoped (i.e. n < 3 x 10 15 cm" 3 ) apart from an
n = 5 x l 0 1 7 c m ~ 3 sublayer 0-5/nn below the lower core
boundary. Table 1 gives other details of the structures. Standard photolithographic procedures were used to make stripELECTRONICS LETTERS 9th April 1987 Vol. 23
No. 8
loaded waveguides by wet-etching ribs in the upper cladding.
Twelve sets (seven widths, 2-10//m, 50 urn apart) of guides
were available per chip.
Discussion of results: A series of waveguide growths incorporating a number of variations is summarised in Table 1, which
also gives the loss measurement results, measured on a chip
about lcm long in each case. The chip was recleaved to
obtain R only for representative growths, C, D, F and G. Fig.
1 displays the derived R values for these cases, while Fig. 2
plots loss against width for the best material (F). The general
variation of loss with width in Fig. 2 typifies all the samples.
However, the slope for narrow guides was sensitive to rib etch
depth; thus the wide guides are taken to be most representative of basic material quality. Owing to the greater uncertainty
in measurement of these, due to overmoding, extrapolation
from the narrow guide results has been used to some degree.
It is not possible to say with certainty that grading the
heteroj unctions reduces guide loss. However, this seems likely
since, out of the five growths on the same substrate type
(C-G), the three with the same grading (C-E) had comparable
losses irrespective of cladding aluminium fraction or symmetry; the one upgraded growth (G) was lossy while F, with
the lowest loss of all, used a more relaxed grading. Substrate
differences partially obscure the remaining results (A, B and
H), which nevertheless suggest that the initial (substrate to
lower cladding) grade may be the most important.
16
CD 1 2
•
.
'
limited sample due to
multimode guiding
o
.2 0 8
"5
[qualitative 1
[trend only]
3
4
5
6
7
8
nominal waveguide width, um |976/2|
Fig. 2 Measured waveguide losses as a function of width for sample F
.
R
= 32
a = 156
2
The spread of R values for each growth is due to the accumulated error of the three independent measurements combined to derive R. Their normalised mean variances can be
shown to sum to the normalised R variance; thus, assuming
the same deviation for all three, we can derive the measurement accuracy from the spread on R. For case F we obtain an
individual loss measurement accuracy of ± 0 1 dB. Equivalent
figures for the other growths of Fig. 1 (entered in Table 1) can
be seen to vary with the loss measured.
R = 29 9
a = 1 28
R = 28 9
.A\
t
Si 1 0
02
growth
reference
a = 108
R = 28
a = 1-94
24
higher-order waveguide modes—by reducing the apparent
cavity finesse give a spuriously high loss figure and (usually)
low derived R. Thus the lowest measured losses may legitimately be given the greatest weight as those least likely to be
in error, and representing guides with fewest defects. To avoid
inaccuracy due to multimode guiding only narrow guide
results (W < 4/xm) are included in Fig. 1. The mean R values
(close to the expected 30% for plane waves) show that the
remaining error mechanisms were not, on average, significant.
With care to preferentially excite the fundamental mode, loss
values could be obtained for the wider guides which were
consistent with extrapolation from the monomode results (e.g.
Fig. 2). However, these should be treated with caution and
regarded as an upper bound.
26
28
30
32
34
36
|97 6/il
facet reflectivity R, •/•
Fig. 1 Scatter of derived facet reflectivity values for some of the growths
of Table 1
Mean and standard deviation are indicated
Measurement accuracy: Two factors which may cause the
effective R to differ from the plane-wave value are (i) the offperpendicular facets, which reduce the coupling between
forward and reverse waveguide modes, and (ii) the inherent
spot size dependence of R due to the nonplanar wave fronts.
Both of these cause R to vary slightly with vertical confinement, and hence withi cladding aluminium fraction and core
thickness. Fig. 1 demonstrates the expected trend.
It is noteworthy that all error mechanisms other than
noise—e.g. facet imperfections, spectral impurity, stray light,
Conclusions: We have demonstrated the lowest losses reported
to date ( < 0-7 d B c m " 1 at 3/xm width) for GaAs/GaAlAs
double heterostructure waveguides designed for efficient highspeed electro-optic devices. The Fabry-Perot loss measurement technique 3 has been shown, by self-consistency, to be
accurate to ± 0-1 dB cm ~ * for lowest losses. This work was
supported in part by the UK Department of Trade &
Industry under the JOERS scheme.
20th February 1987
R. G. WALKER
H. E. SHEPHARD
R. R. BRADLEY
Plessey Research Caswell Limited
Caswell
Towcester, Northants. NN12 8EQ, United Kingdom
References
1
H1RUMA, K., INOUE, H., ISHIDA, K., a n d MATSUMURA, H.: ' L o w l o s s
GaAs optical waveguides grown by the metal organic chemical
Table 1 WAVEGUIDE LAYER STRUCTURES AND SUMMARY OF MEASUREMENT RESULTS
Growth
ref.
Substrate
Aluminium fractions
(lower/core/upper)
mole %
A
B
C
D
E
F
G
H
CZn ++
CZ«
CZSI
CZSI
CZSI
CZSI
CZSI
Boat n+
/ 15:0: 25
/ 15:0: 25
: 15:0:25
: 15:0: 15
: 10:0:10
: 10 : 0:10
/ 10/0/10
: 10 : 0 : 10
Boundary grading
distance
core
lower
Approx.
core
thickness
Mean
derived
R
R
used
%
—
—
3205
29-88
—
28-86
2800
—
%
32
32
32
30
29
29
29
29
lixa
fan
/an
0
0
10
0-5
0-5
0-5
0
0-5
01
01
01
01
01
0-2
0
01
0-9
0-9
0-9
0-9
10
10
10
10
Minimum loss
4^m
Sfitn
width
width
dBcm"'
dBcm- 1
1-8
3-4
0-9
0-85
06
0-5
3-5
1-25
1-5
2-9
0-55
0-65
0-55
0-3
3-2
0-9
Derived
accuracy
dBcm"1
—
—
+ 013
±0115
—
±01
±02
—
Key: / = abrupt, : = graded
ELECTRONICS LETTERS 9th April 1987
Vol. 23
No. 8
363
vapour deposition method', Appl. Phys. Lett., 1985, 47, pp.
186-187
LIN, S. H., WANG, S. Y., NEWTON, S. A., and HOUNG, Y. M.: 'Low-loSS
GaAs/GaAlAs strip-loaded waveguides with high coupling efficiency to single-mode fibres', Electron. Lett., 1985,21, pp. 597-598
WALKER, R. G.: 'Simple and accurate loss measurement technique
for semiconductor optical waveguides', ibid., 1985, 21, pp. 581—
583; Erratum: ibid., 1985, 21, p. 714
ELECTROMIGRATION EFFECTS IN POWER
MESFET RECTIFYING AND OHMIC
CONTACTS
Indexing terms: Semiconductor devices and materials, FETs,
Power semiconductor devices, Reliability
Both gate and source/drain electromigration are significant
failure mechanisms in power MESFETs. The correlation
between electromigration effects due to high current density
and measured electrical degradation is investigated in devices
of different technologies. A safety zone of operation for
ohmic contact electromigration is defined.
Power MESFETs for applications in microwave communications systems must guarantee good operational reliability. In
RF and large-signal conditions these devices could operate
with high pulsed current density through the gate rectifying
contact. Moreover, DC polarisation current may induce
similar high-density values in the source/drain contacts. Electromigration of both gate and source/drain metallisations may
be an important degradation mechanism, as has often been
reported in the literature.1"5 In any case, until now little
experimental evidence of this phenomenon has been offered.
In the framework of an extensive evaluation of power
MESFET reliability, we have investigated the correlation
between electromigration and electrical degradation, by performing DC accelerated tests on commercial devices of different technologies. In particular, we arranged a high forward gate
current (HFGC) test at Tch = 200°C and J = 5 x 10s A/cm2
to investigate gate electromigration, while we used DC life
tests at different temperatures up to Tch — 240°C to stress electromigration effects in ohmic contacts. In this letter we report
the most significant result obtained in both tests.
Gate electromigration: During HFGC testing we forced a high
forward current density of 5 x 105A/cm2 while keeping the
ohmic contact current density low. This situation must be
4 pm
considered as highly accelerated with respect to real operating
conditions, in which these high values can be attained only
during short pulses.6 This gate current density caused electromigration phenomena even after only lOOOh of HFGC testing
for Al-gate fingers. As shown in Fig. 1, the Al-finger interruption can be clearly observed with both secondary electron
SEM (Fig. la) and electron-beam-induced current (EBIC)
techniques (Fig. Ib). However, very thin interruptions are
more evident with EBIC analysis. More often than not the
interruption occurs at the beginning of the finger, where the
current density is higher7 and the metallisation could be
thinner at the mesa step. From the electrical point of view,
this phenomenon prevents pinch-off being achieved and
increases the gate series resistance. In real operating conditions, the loss of gate control substantially degrades the characteristics of the MESFET, increases signal distortion and
eventually causes the catastrophic breakdown of devices.
Drain/source electromigration: All tested devices have alloyed
gold-germanium-based ohmic contacts, subsequently thickened either by gate metallisation deposition (when Au-based:
this technology is commonly used in medium-power
MESFETs up to 0-5 W), or by electrolytic gold growth.
alloyed
ohmic
contacts
20pm
IS
thickened
gold
te
1954/21
Fig. 2
a Schematic diagram of electron wind along drain finger
b, c SEM images of two drain contacts for devices that endured
5000 h of life testing at Tch = 200°C and J = 5-3 x 10s A/cm2
During DC life tests, in medium-power MESFETs we
observed Au electromigration phenomena at the drain contacts. The photographs in Figs. 2b and c show SEM images of
two drain contacts for devices that endured 5000 h of life
testing at Tch = 200°C and J = 5-3 x 105 A/cm2. Both gold
removal at the end of the fingers and accumulation at the
beginning are evident, in agreement with theflowof the electron wind, as indicated in Fig. 2a. This effect seems to be
much clearer than those reported in the literature thus far.1'2
230
220
drain
u
-
-210
Qj
2 200
<h
Q.
I 190
t 170
1954/H
a Secondary electron SEM image of Al-gatefingerbreakage due to
electromigration, after 1000 h of HFGC test at Tch = 200°C and
J = 5 x 105 A/cm2
b EBIC image of same finger
364
10"*
10"
10°
drain current density , A/cm 2
10'
Fig. 3 Channel temperature against drain current density for different
devices
Ohmic contact electromigration, after 10000 h of testing, can be
observed only in stress condition marked ' C in Figure
ELECTRONICS LETTERS 9th April 1987 Vol. 23 No. 8
Документ
Категория
Без категории
Просмотров
0
Размер файла
527 Кб
Теги
3a19870266
1/--страниц
Пожаловаться на содержимое документа