IEEE JOURNAL OF QUANTUM ELECTRONICS, VOL. QE-21, NO. …klau/pubs/03.pdfIEEE JOURNAL OF QUANTUM ELECTRONICS, VOL.QE-21, NO. 2, FEBRUARY 1985 121 Papers Ultra-High Speed Semiconductor

IEEE JOURNAL OF QUANTUM ELECTRONICS, VOL. QE-21, NO. 2, FEBRUARY 1985 121

Papers

Ultra-High Speed Semiconductor Lasers

Abstract-Recent progress on semiconductor lasers having a very high direct modulation bandwidth of beyond lOGHz willbe described. Issues related to application of these lasers in actual systems will be addressed. Possibilities of further extending the bandwidth of semiconductor lasers will be examined.

S I. INTRODUCTION

EMlCONDUCTOR lasers are devices of great importance for optical communication systems due to their small size,

high efficiency, and high speed for direct modulation. The direct modulation of semiconductor lasers has been a subject of active research for the past 20 years. Experiments have shown the existence of a resonance peak in the modulation response, whose frequency varies as the one-half power of the injection current above the lasing threshold. These results were well explained by a simple model based on the dynamic laser rate equations [l]

dN J d t ed r,

lV. A(N - Nom) P

where N is the carrier density, Nom is the carrier density for transparency, P is the photon density in a mode of the laser cavity, J is the pump current density, d i s the thickness of the active layer, 7 , is the spontaneous recombination lifetime of the carriers, 7 p is the photon lifetime of the cavity, o( is the optical gain coefficient, p is the fraction of spontaneous emission entering the lasing mode, and e is the electronic charge. Equation (1 j is simply a bookkeeping of the supply, annihila- tion, and creation of carriers and photons inside the laser cavity and describes laser dynamics in a most basic manner. More detailed explanations of laser modulation behaviors can be obtained by addition and/or modification of terms in (1). In the early stages of semiconductor laser development, the principal concern was in optimizing laser structures for improvements in the static lasing characteristics in terms of

Manuscript received July 16, 1984; revised October 5, 1984. This work was supported by the Defense Advanced Research Projects Agency and by the Naval Research Laboratory.

K. Y . Lau is with Ortel Corporation, Alhambra, CA 91803. A. Yariv is with the Department of Electrical Engineering and Applied

Physics, California Institute of Technology, Pasadena, CA 91125.

threshold current, quantum efficiency, linearity of light versus current characteristic, operation at high optical power, and long-term reliability. As the state of semiconductor laser technology approaches maturity, their high-speed dynamic characteristics become a subject of increasing importance. For many years the practical direct modulation bandwidth of semiconductor lasers has been commonly accepted to be in the lower GHz range ( 5 2 GHzj. The work of Figueroa et al. [2] was one of the first major efforts to tap the high frequency potential of high quality commercial semiconductor lasers. Their work has shown that a modulation bandwidth of 4-5 GHz is possible in a number of common laser structures. How- ever, achieving that kind of modulation bandwidth requires that the lasers be biased at a level well above their nominal ratings, dangerously close to the point of catastrophic damage. This paper will describe recent research efforts which have led to the development of advanced laser structures possessing a direct modulatjon bandwidth of beyond 10 GHz under reliable room temperature continuous operation. We will begin by considering, in the next section, the theoretical groundwork which identifies fundamental laser parameters of direct physical rele- vance that affect the modulation bandwidth of injection lasers.

11. RELEVANT PHYSICAL PARAMETERS A considerable amount of information concerning the dynam-

ical behavior of injection lasers can be gained by a small signal analysis of the rate equations-a procedure that linearizes (1). The resulting modulation response has the transfer characteristic of a second-order low-pass network [3] . A resonance in the modulation response, known as relaxation oscillation in lasers, results from the interplay between the optical field and the population inversion. The strength of the resonance depends, among other things, on the spontaneous emission factor, lateral carrier diffusion, and presence of saturable, absorb- ing defects. The relaxation oscillation frequency obtained from the small signal analysis is

where rp is the photon lifetime, given by rp = l/u(a + 1/L In 1IR) where u is the group velocity of the light, 01 is the distributed loss, L is the length of the cavity, and R is the mirror reflectivity; p o is the steady-state photon density and A is the differential optical gain constant as given in (1). The modula-

0018-9197/85/0200-0121$1.00 0.1985 IEEE

122 IEEE JOURNAL OF QUANTUM ELECTRONICS, VOL. QE-21, NO. 2, FEBRUARY 1985

tion bandwidth of the laser is widely accepted to be equal to f,. Equation (2) expresses the modulation bandwidth as a simple function of three fundamental, independent parameters of direct physical significance. The differential optical gain constant A is governed directly by material properties. The photon lifetime rp can be linked directly to device geometry. The photon density is perhaps the single most important parameter expressing the state of a laser, since it is directly responsible for many other important laser parameters such as intrinsic single mode linewidth, relative intensity noise, and catastrophic facet damage. It is interesting to compare (2) with the expression [4] , [ 5 ]

used extensively in the past to predict the modulation bandwidth of semiconductor lasers. The failure of (3) to explain the experimental data has been noted by a number of workers [6], [7]. Paoli [8] showed that the bandwidthisindeedlarger than indicated by (3) and that much better agreement is obtained by replacing r p , the photon lifetime, with the much shorter lifetime that accounts for absorption losses in the unpumped laser. The source of the difficulty with (3) can be appreciated if we rewrite (2) in an equivalent form:

where Jth is the threshold current density and is the optical confinement factor. We see that if one assumes I’No,Arp << 1 then we obtain (3). The above approximation amounts to neglecting the fact that in a semiconductor laser a large electron density No, is needed just to achieve transparency. As a matter of fact, in a semiconductor laser, FN0,Arp is of the order of magnitude of 4-5 so that we need to use the exact expression (4), or equivalently, the simpler and more fundamental expression (2).

Equation (2) suggests three obvious ways to increase the relaxation frequency-by increasing the optical gain coefficient or the photon density, or by decreasing the photon lifetime. The gain coefficient A can be increased roughly by a factor of five by cooling the laser from room temperature to 77 K [9] . Biasing the laser at higher currents would increase the photon density in the active region, which simultaneously increases the optical output power density rout according to

1 2

Iout = - p o h wu In

where hw is the photon energy, u is the optical group velocity, and R is the end mirror reflectivity. Catastrophic mirror damage occurs at a power density of rout, cat = 1 MW/cmZ for a laser with a mirror reflectivity of R = 0.3 [ lo] . This sets an upper limit on the maximum permissible photon density and, hence, the maximum modulation bandwidth. This limit can be increased very considerably by using a window structure.

The third way to increase the modulation bandwidth is to reduce the photon lifetime by decreasing the length of the laser cavity. Such a laser has to be driven at higher current

densities, and thermal effects due to excessive heating will limit the maximum attainable modulation bandwidth. To illustrate these points, the relaxation frequency as a function of the cavity length and the pump current density is plotted in Fig. l(a) using (2) and the static solutions of (1). Also plotted in Fig. 1 is the power density at the mirror using (5). As an example, a common laser with a cavity length of 300 pm operating at an output optical power density of 0.8 MW/cm-2 possesses a bandwidth of 5.5 GHz, and the corresponding pump current density is 3 kA/cm2. Operating at an identical power density, the ’ bandwidth is 8 GHz for a shorter laser with a cavity length of 100 pm, but the corresponding current density is 6 kA/cm2. A higher current density alone may not be a cause for rapid degradation of lasers. To support thisassertion, we note that lasers with increased optical damage threshold, as described above, can operate at increased current densities without appreciable degradation of their reliability [I I ] . Fig. l(b) shows plots of the same quantities as in Fig. l(a) but for a laser operating at liquid nitrogen temperature. The increase in bandwidth is a direct result of the increase in A . It can be seen that a modulation bandwidth beyond 20 GHz can be achieved; however, incorporation of a short optical cavity and/or a window structure is imperative under these operating conditions.

The above discussions are based on small signal analysis and are therefore strictly valid only in this regime. The small signal regime usually refers to cases in which the laser is biased above threshold, and the modulation amplitude in the optical output is a “small” fraction of the dc bias level. In reality, that “small” fraction can be as big as 70 percent. Although injection lasers behave differently under large and small signal modulation, experimental and theoretical results in the small signal modulation regime are often good indicators of the large signal performance when the laser i s biased above threshold, as in digital pulse-code modulation of a laser diode. The bulk of the discussions in the following sections are mainly concerned with the small signal regime. In Section XII, an illustration of the transition from small signal to large signal regime will be given.

111. SHORT-CAVITY LASERS Experiments have been performed to determine the modula-

tion bandwidth achievable in a short-cavity laser. The lasers used were buried heterostructure lasers fabricated on a semiinsulating substrate (BH on SI) [12] , a schematic diagram of which is shown in Fig. 2. In addition to a low lasing threshold (typically <15 mA), which is necessary to avoid excessive heating when operated high above threshold, these lasers possess very low parasitic capacitance [I31 , which otherwise would obscure modulation effects at high frequencies ( 2 5 GHz). The standard measurement system, shown in Fig. 3, consists of a sweep oscillator (Hp8350) used in conjunction with a network analyzer (Hp8410 series) and a microwave s-parameter test-set (Hp8746B). The photodiode used was a high speed GaAs pin diode fabricated on semiinsulating substrate. The photodiode was an improved version of’the one reported previously [ 141 ; its frequency response was calibrated from dc to 15 GHz using a mode-locked dye laser and a microwave spectrum analyzer. The -3 dB point of the photodiode response is

LAU AND YARIV: ULTRA-HIGH SPEED SEMICONDUCTOR LASERS 123

I 2 3 4 5 6 7 8 9 1 0 CURRENT DENSITY [kA/cm']

(a)

CURRENT DENSITY [kA/crn']

(b) Fig. 1. (a) Relaxation frequency ure1 (solid lines) and optical power

density outside the mirrors (dashed lines) as a function of the cavity length and pump current density at T = 300 K. The following parameters are used: active layer thickness = 0.15 gm, 01 = 40 cm-', R = 0.3, u = 8 x lo9 cm . s-l, A = 2.56 X cm3s-', r = 0.5, Nom = 1 X l0 l s ~ m - ~ , B = 1.5 X cm3 s-', Fzw = 1.5 eV. (b) same as (a) but at T = 77 K. The same parameters as in (a) are used. except: A = 1.45 X cm3sL', Nom = 0.6 X lo1' ~ m - ~ , B = 11 X cm3s-' [5]. B is the bimolecular recombination constant of the spontaneous recombination of carriers.

Fig. 2. Schematic diagram of a buried heterostructure laser fabricated on semiinsulating GaAs substrate.

SWEEP OSCILLATOR

MEASUREMENT (TRANSFER CHARACTERISTIC FROM PORT I TO PORT 2 )

PORT I

I 4 Fig. 3. Diagram of the standard high frequency laser measurement set-

up used in this work.

at 7 GHz and the -5 dB point at 12 GEIz. The observed modulation response of the laser is normalized by the photodiode response at each frequency. The modulation responses of a 120 pm long BH on SI laser at various bias levels are shown in Fig. 4. The modulation bandwidth can be pushed to beyond 8 GHz by operating near the catastrophic damage point. Fig. 5 shows the relaxation oscillation frequency of this laser as a function of 6 where Po is the output optical power, together with that of similar lasers with longer cavity lengths. The advantage of a short-cavity laser in high frequency modulation is evident.

It must be emphasized that in order to take advantage of the reduced lifetime, a very low threshold laser is required for practical realization of these short cavity lasers so that 1) catastrophic damage, not heating, is the basic limitation and 2) the increased current density has no significant effect on device reliability. As one shortens the laser cavity the current density will continue to rise, until a point is reached beyond which long-term reliability is conpromised. Life-test data of commercial BH on SI lasers have shown that, at least for the laser structure considered above, that point occurs at L 150 pm.

IV. Low TEMPERATURE OPERATION In this section, we describe experimental results on direct

amplitude modulation of low threshold BH on SI lasers operating below room temperature. This led to the first demonstra- tion of a direct modulation bandwidth of beyond 10 GHz in a semiconductor laser.

The lasing threshold current of a 175 pm long BH on SI laser at room temperature is 6 mA, dropping to -2 mA at -70°C. The I-V curves reveal a drastic increase in the series resistance of the laser below -60°C. This is believed to be due to carrier freeze-out at low temperatures since the dopants being used, Sn (n-type) and Ge (p-type) in GaAlAs, have relatively large ionization energies. Modulation of the laser diode becomes very inefficient as soon as freeze-out occurs, because of a reduction in the amplitude of the modulation current due to a higher series resistance.

The frequency response of the lasers was measured using the standard arrangement. Fig. 6 shows the response of a 175 pm long laser at - 50"C, at various bias levels. The relaxation resonance is quite prominent at low optical power levels. As the optical power is increased, the resonance gradually subsides, giving way to a flat overall response. The modulation bandwidth, taken to be the corner frequency of the response (the frequency at the relaxation resonance peak or at the -3 dB point in cases when it is absent), is plotted against the square root of the emitted optical power (6) in Fig. 7, at room temperature and at - 50 and - 70°C. The relative slopes of the plots in Fig. 7 yield values for the relative change in A as the temperature is varied. The ratio of the slope at 22°C to that at -50°C is 1.34 according to Fig. 7. This factor is fairly consistent (between 1.3 and 1.4) among all the lasers tested, even including those from different wafers. According to these measurements we deduce that the intrinsic differential optical gain increases by a factor of -1.8 by cooling from 22 to -50°C. This is consistent with theoretical gain calculations [ 151 .


Frequency

Fig. 4. Modulation characteristics of short cavity (120 pm) BH on SI laser at bias power levels of (a) 1 mW, (b) 2 mW, (c) 2.7 mW, and (d) 5 mW.

01 I I I I 0 0.5 1.3 1.5 2.0 2.5

( rnW1'2)

Fig. 5. Measured relaxation oscillation resonance frequency of lasers of various cavity lengths as a function of a, wherePo is the CW output optical power. The points of catastrophic damage are indicated by downward pointing arrows.

I 2 3 4 5 6 7 8 1 0 1 2 1 4 Frequency (GHz)

Fig. 6. Modulation response of a 175 pm buried heterostructure laser on semiinsulating substrate operating at -50°C.

V. HIGH PHOTON-DENSITY DEVICES Although the results of the previous sections demonstrated

that a direct modulation bandwidth of beyond 10 GHz can be accomplished by lowering the operating temperature of the laser to - 50°, it is more desirable to attain the bandwidth under room temperature operation. This can be accomplished by

Fig. 7. Variation of modulation bandwidth (corner frequency of the

power f l . modulation response) with the square root of the emitted optical

tackling the third parameter in (2), the optical power density Po. As pointed out above, an increase of the optical power can bring about undue degradation or even catastrophic damage to the laser unless the structure of the laser is suitably designed. One common means of raising the ceiling of the reliable operating power of semiconductor lasers is by means of a large optical cavity [16] . The mechanism responsible for a higher catastrophic damage power in these devices is by lowering the optical power density at the active layer, since such damage commonly originates from the active layer near the crystal facet. This maneuver, however, serves little to increase the modulation bandwidth because the quantity of concern here, the photon density within the active region [ p o in (2)] , re- mains unchanged. A laser suitable for high-speed operation should therefore be one with a tight optical confinement in the active region along the entire length of the laser, with a transparent window at the end regions able to withstand much larger photon densities without catastrophic damage. The use of a transparent window structure to increase the catastrophic damage level has previously been demonstrated [17] , [18] . Using a window buried heterostructure laser fabricated on a semiinsulating substrate, we have demonstrated for the first time a direct modulation bandwidth exceeding 10 GHz for a semiconductor laser operating at room temperature.

The laser used in this experiment is shown in Fig. 8. The device is structurally similar to'the BH on SI laser (Fig. 2) except that here the end regions near the facets are covered by a layer of unpumped GaAlAs which forms a transparent window. The laser was fabricated using a two-step liquid phase epitaxial (LPE) growth process, followed by mesa stripe etching on which part of this stripe was removed in order to grow the


Fig. 8. Schematic diagram of the window buried heterostructure laser on semiinsulating substrate.

GaAlAs windows. In the second LPE growth two blocking GaAlAs layers were grown [19] . Individual devices were fahri- cated using a precise cleaving technique in the window regions within several microns from the edge of the double heterostructure. The optical wave propagates freely in the end window region. As a result of diffraction, only a small amount of light reflected from the crystal facet couples back into the active region. The theoretical value of the effective reflectivity, assuming a fundamental Gaussian beam profile, is -5 percent for L = 5 pm. The actual values of L for the devices fabricated lie around this value.

The threshold current of window BH on SI lasers ranges from 14 to 25 mA. The threshold current of comparable BH on SI lasers ranges from 10 to 20 mA, the difference being due to the reduced reflectivity of the window region. The difference is small because the amount of pump current required for overcoming mirror loss is only a small fraction of that for bringing the lasing medium to optical transparency. The catastrophic damage threshold in these devices is beyond 120 mW under pulse operation. Under CW operation, the maximum operating power is limited by heating to 50 mW, The normalized modulation response of a window laser is shown in Fig. 9; at various bias optical power levels. The conspicuous absence of the relaxation oscillation peak should be constrasted with the responses of similar devices which are capabie of being modulated to comparably high frequencies (-10 GHz) as described previously. This is due to a superluminescent damping effect [20], [21] .as a result of the reduced facet reflectivity due to presence of the window. A plot of the -3 dB modulation bandwidth of the window BH on SI laser against the square root of the bias optical power is shown in Fig. 9(b). Contribu- tions from parasitic elements are believed to be at least partly responsible for the departure of the observed data from a linear relationship at high frequencies. Also shown in Fig. 9(b) are the bandwidth characteristics of a large optical cavity BH on SI laser. The substantially lower modulation speed at comparable output power levels bears witness to the fact that the optical power density at the active region of these devices is lower than that in lasers of conventional construction.

The results described in this section, together with those of the last two sections, demonstrate the dependence of the modulation bandwidth on three fundamental laser parameters as given in (2). With these results in hand, it is quite conceiv- able that the direct modulation bandwidth of semiconductor lasers can be extended to the 20 GHz range by optimizing all of the three parameters simultaneously according to the theoretical results in (2). However, as the results in this and the

Modulation Frequency

(a)

I I I I I 1 ;'*I I

'3 10

f i 0 (rnW''')

(b) Fig. 9. (a) Modulation characteristics of a window buried heterostruc-

ture laser on semiinsulating substrate at various bias optical power levels at room temperature. The curves 1-5 correspond to bias optical powers of 1.7, 3.6, 6.7, 8.4, and 16 mW. (b) The -3 dB modulation bandwidth versus the square root of the emitted optical power. The length of the laser is 300 Elm.

last section show, one has still to reckon with the problem of laser parasitic elements before such bandwidths can be realized. The subject of parasitic elements will be discussed in the next section.

VI. PARASITIC ELEMENTS LIMITATION A common source of concern when dealing with high fre-

quency electronics is the adverse effects that parasitic elements might have on device performance. Modulation of semiconductor laser diodes at high frequencies is no exception. Are the observed modulation characteristics due to the laser alone or are they due to parasitic elements? Wili bond-wires limit the frequency response of laser diodes? Does one laser structure possess a larger parasitic capacitance than others? To answer these questions one must treat the laser as an electrical element and establish an equivalent circuit for it, together with the parasitic elements. Characterization of networks at high frequencies is most commonly done with s-parameters ("s" stands for scattering) [ 2 2 ] . The ease with which these parameters are measured makes them especially suited for describing microwave devices. s-parameters can be easily converted to other, more familiar descriptions of the network: when a Smith chart is laid over a polar display of sI1 or sz2 , for example, the input or output impedance can be read directly. If a swept-


frequency source is used, the display becomes a graph of input or output impedance versus frequency [23].

Parasitic elements aside, let us first consider the equivalent circuit of the intrinsic laser diode. The elements of the laser equivalent circuit are derived from the rate equations (2), which describe the interplay between the optical intensity (or photon density) and the injected carriers. These equations are aug- mented by the voltage-carrier density characteristics of the injection junction. The resulting equivalent circuit is a parallel RLC resonance circuit. The magnitude of the impedance of the entire equivalent circuit I Z(w)l is essentially zero at all frequencies except near the relaxation oscillation resonance frequency, where its value does not exceed -1 8. This reflects the fact that gain clamping at these frequencies is manifested in the equivalent circuit as an ac short so that no ac voltage, and hence Fermi-level fluctuations, can develop. Consequently, in comparison to relatively large external elements, the intrinsic laser diode can be regarded as a short circuit at all frequencies (as long as the laser is biased above threshold).

Fig. 10 shows the general geometry of a laser fabricated on a conductive substrate. A simple equivalent circuit of this laser is shown at the bottom of the figure. It includes the major parasitic elements involved. One can also include a submount capacitance by connecting a shunt capacitance to the entire circuit in Fig. 10. For a well-constructed mount, the value of this capacitance is usually 5 0.2 pF and can be neglected. The value of the bond wire inductance L depends not only on the length of the wire, but also on its diameter and its proximity to the ground plane 12.51. A good wire-bond should not con- stitute an inductance much larger than 1 nH. The parasitic capacitance involved is the capacitance between the top and bottom contact in the area outside the lasing region, and obviously depends on the intervening material, the separation between the contacts, and their area. The value of this capacitance, as we shall see below, varies widely from laser to laser. As it turns out, this capacitance is the major element of concern at high frequencies.

Before discussing how the values of the parasitic elements depend on laser structure, we shall first look at how the para- sitics affect the modulation response of injection lasers. We must now take into account how the laser is being driven. In usual experiments where a laser is mounted at the end of a 50 8 transmission line, and no provision is made for impedance matching, reflections of the microwave drive signal from the laser must be taken into account. This is equivalent to driving the laser with a pure voltage signal source having a source impedance of 50 a. In an experiment where a network analyzer is used to measure the response of the laser, the ac signal from the photodiode is compared to the microwave signal feeding the laser. This microwave feed signal is represented by a pure voltage source. The overall modulation response is then obtained by multiplying the intrinsic response of the laser by the following ratio :

current flowing into the intrinsic diode I voltage of the signal source

The manner in which 1) is defined takes into account electrical reflection due to impedance mismatch of the laser in a 50 8

oxlde P GaAs

p GaAlAs p GaAs N GaAlAs N Ga As

- -

14 /bond wlre Inductonce i contact resistance

‘lntrlnslc dlode

Fig. 10. General geometry of a laser fabricated on a conductive substrate and a simple equivalent circuit.

system. An analysis of the equivalent circuit in Fig. 10, under a 50 8 drive, gives the following for Q:

1 1)- ( iwy iw

- - T + - + I

where

and L, R , and C are the parasitic components as shown in Fig. 10. The response of (6b) is in the form of a second-order low- pass filter type. For reasonable values of L , R , and C, the Q of this response is close to or less than 3, which means that no resonance occurs in the injection ratio Q at w = wo. It is intuitively obvious, by observing the equivalent circuit of Fig. 10, that if the inductance L is “verysmall,” then the principal effect of the parasitic elements comes from the RClow-passcombina- tion, where the presence of the capacitor bypasses an increasing amount of drive current at high frequencies. We shall illustrate by the following examples the range of value of L that are to be considered “very small.” First, we take the following values for the parasitic elements: L = 1 nH, C = 4 pF, and R = 4 8 (which, as we shall illustrate later on, are typical values for a BH on SI laser with a fairly long (-1.5 mm) bond wire). The values of wo and Q are [see (6c), (6d)l 9.25 GHz and 0.516, respectively. A plot of the injection ratio versus frequency is shown in Fig. 11 (labeled case A in the figure). Next, we take C= 4 pF and R = 4 8 as before, but let L = 0.2 nH (which corresponds to a short bond wire -0.3 mm in length). The values of w o and Q in this case are 20.7 GHz and 0.416, respectively. The injection ratio in this case is plotted in Fig. 11 (case B). There are now two distinct corner frequencies, at 11.1 and 38.7 GHz. This is what is commonly referred to as an underdamped second-order filter response, and arises from


I I I 2 4 6 8 1 0 20 30 40 50

Frequency (GHz)

Fig. 11. The current injection ratio q as a function of frequency, for two cases with the values of the parasitic elements as shown on the figure.

a splitting of the poles on the negative real axis. Physically, this is just the case we have mentioned above, that L is suffi- ciently small so that the Recombination dominates (which produces the rolloff at the first corner frequency at 11.1 GHz). As mentioned before, a good wire bond in a well-designed high- frequency package should not contribute more than 1 nH of inductance, with typical values of 0.2-0.3 nH, and therefore, the inductance is really of no concern. The R C combination is, in practice, the principal limiting factor in high-speed modulation at frequencies up to at least 15 GHz. In some lasers where the parasitic capacitance was measured to be as large as 40 pF, a dip in the modulation response in the lower GHz range is expected and has been observed by numerous researchers

Experimentally, the value of the parasitic elements can be determined by measuring the microwave s-parameter sI1 (reflection coefficient) at various frequencies and then fitting the data to the calculated reflection coefficient of the assumed network. A satisfactory fit cannot usually be obtained at all frequencies unless the assumed circuit is reasonably similar to the actual circuit (i.e., it contains all the major elements in the right topology). Extensive s-parameter measurements reveal that the equivalent circuit in Fig. 10 produces reasonable fits to most lasers fabricated on a conductive substrate [29] , [30] . Recently, Tucker [31] measured parasitic elements in various types of laser and correlated the results with the actual modulation performance of the lasers.

A separate class of lasers can be defined whch are based on semiinsulating substrates instead of conventional n”-doped substrates [32] , [33]. The buried heterostructure laser fabricated on semiinsulating substrate, which is the “basic” laser structure used in most of our high frequency experiments, is shown again in Fig. 12, together with its equivalent circuit. The major difference betweeh the equivalent circuit of this laser and conventional lasers on conductive substrate is that the parasitic capacitance in the BH on SI laser is in the form of a distributed network, owing to the particular geometry of this device. The resistance of the bottom n” contact layer provides an effective shielding of the capacitances far away from the junction,

[26] -[30].

Burled-Heterostructure Loser on SI Substrate

SI Substrate

. . T J T - T T T Dlstrlbuted Copocitance

Fig. 12. Parasitic elements in a BH laser on semiinsulating substrate.

especially at high frequencies. Assume that the resistance per unit length of the bottoin n” layer is Rdist and the parasitic capacitance per unit length between the top contact and the bottom n+ layer is,Caist. Then, the effective impedance of the entire distributed RC ladder network is given by analogy to transmission line theory:

where W is the width of the top contact,Z, is the characteristic impedance of the ladder network

and k is the “propagation constant”

k = &WRdist Cdist.

At very high frequencies, the real part of k is large, with the physical significance that the electric field does not penetrate too far beyond the lasing junction. Capacitance elements exist- ing far away from the junction therefore do not contribute to the effective total capacitance. The total impedance of the distributed network is 2, in this limit. The effect of distributed capacitance can be understood in a more intuitive manner by considering the limit of small Rdkt and expanding (7) in this limit :

2 = 2, coth (kW)

where n

Ceff = L

1 + W2R 2C2

45

R 2 R e f f = - - -

3 945 u 2 R 3 C 2

and R and C are respectively Raist W and W , the total para-


~ X I R Cm=0.O7pF

L = 0.18 n H T ~ / R C, = 0.05 pF

L = 1.0nH

R = { 2 forward bias reverse bias

C,= 4.1 pF

R l =0 .5R

R = 2.5n forward blas

Cp=3.0pF

R , = l . O n

03 reverse bias

Sll parameter plot Frequency I GH2 - 8 GH2 In steps o f I GHz

0 theoretical + expetlmental

0 Frequency I GH2 + 8 GHz in steps of I GHz Sll parameter plot

0 theoretical +experimental

forward bias forward blas

I I I I I I I I I ,

0 ) (C) Fig. 13. (a) Effective equivalent circuit of the parasitic elements in a BH on SI laser. (b), (c) s11 parameter of BH on SI

lasers and theoretical fit to the equivalent circuit shown on the top of the figures. Forward bias: a 250 pm long laser with a short (<0.3 mm) bond wire; reverse bias: a 175 pm long laser with a relatively long (1.0 mm) bond wire.

sitic capacitance and lateral resistance of the bottom conductive n-layer, and only the first four terms in the Taylor expansion of (7) are included. It follows that the equivalent circuit of a laser on SI substrate is that shown in Fig. 13(a). The effective capacitance Ceff thus decreases at high frequencies and the existence of R,ff further reduces the amount of current diverted into the parasitic capacitance at high frequencies. Actual sl l measurements of BH on SI lasers indeed confirm the equivalent circuit of the configuration shown in Fig. 13(a). The magnitude and phase of the sI1 parameter is usually plotted in polar form, with the frequency as the varying parameter. The plot is superimposed on a Smith chart, from which the total impedance of the network under test can be read off at any frequency from the value of sl , a t that frequency. Fig. 13(b) and (c) shows such plots of the experimentally measured data (crosses) and theoretical fit (square) of a BH on SI laser of cavity lengths of 175 ,and 250 pm, respectively, at forward bias and reverse bias conditions. The values of the circuit elements used to produce the theoretical fits are also shown in Fig. 13. The values of the effective capacitance and resistances are excep- tionally low in these devices, with an equivalent RCtime constant of <IO ps, which accounts for the total absence of the dip in the modulation response of these devices even at frequencies approaching 10 GHz, as shown in the results described in the last few sections. The -3 dB cutoff frequency in the drive current into the laser due to the parasitic elements with values shown in Fig. 13 is -12 GHz. This can account for the fact that the observed bandwidth of the lasers cannot be pushed significantly beyond 12 GHz despite the various techniques

which are used to enhance the bandwidth. It is therefore necessary to reduce the parasitic elements by suitable electrode design in order to achieve further improvements in bandwidth. Several possible schemes toward this end are described in the next section.

VII. FURTHER BANDWIDTH ENHANCEMENT Of the three fundamental parameters that govern the modula-

tion bandwidth of injection lasers, (2), the photon density p o , and the photon lifetimes T~ can be tackled in relatively straight- forward manners as discussed in Sections I11 and V. The differential optical gain constant A can be increased, as described in Section IV, by the. somewhat inconvenient way of lowering the operating temperature of the laser. It would be much preferable to devise mechanisms for increasing A by built-in struc- tural or material design in the laser. Two examples of such mechanisms will be discussed in this section, first, by detuned coupled cavity loading of a semiconductor laser, and second, through the use of quantum well or quantum wire laser structures. However, one should realize that these measures will be of little help unless the parasitic limitation described in the last section is overcome.

The concept of detuned cavity loading proposed by Vahala and Yariv [34] and its effect on the modulation response can be visualized quite intuitively as follows. Consider the coupled cavity configuration as shown in Fig. 14, where the semiconductor laser is coupled to an external cavity whose length is comparable to that of the laser cavity. The end facet and the external mirror combined produces in effect a frequency selec-


Detuned loodina

u Effective

mirror reflectivity Ref,

Ref f t I

“Natural“ lasing frequency

Fig. 14. Interpretation of a coupled cavity laser as a simple laser having a wavelength dependent facet reflectivity.

tive mirror, whose reflectivity as a function of wavelength is shown schematically in Fig. 14. Consider a case where, under CW operation, the laser oscillates in a single frequency which lies on the rising slope of the effective reflectivity curve (Fig. 14). It is well known that as one modulates a semiconductor laser, considerable wavelength-chirping occurs due to the relatively large variation of the refractive index with carrier density. Therefore, as one modulates a semiconductor laser which is detuned loaded, the frequency chirping would result in ‘an effective increase in the differential gain (or equivalently, effective decrease in the differential loss) if’the steady-state lasing wavelength sits on the side of rising slope as illustrated in Fig. 14. The direct modulation bandwidth ‘increases according to (2). It turns out that this effect also’causes the intrinsic linewidth of the laser to decrease as;wcreQ [34] . It is obvious that the opposite effect would hold if the steqdy-state lasing wavelength sits on the falling edge of the effective reflectivity curve.

The above argument is, of course, only heuristic. * A detailed field analysis which involves a ’ soiution of the coupled equations of motion for the complex field amplitude in each reso- nator bears out the above prediction in a rigorous manner. Fig. 15 shows the calculated variaiiqn of the effective differential gain versus 6 = 2 ~ ~ ( 0 - 0,) where rm is the photon lifetime of the external cavity, w is the actual lasing frequency, and w, is a resonance frequency of the external cavity. This plot indicates that, under slight positive detuning, direct modulation bandwidth enhancement of 2X is possible. It turns out that FM suppression and phase noise reduction are achieved simultaneously under this arrangement. All of these improvements result from the mixing of differential gain and differential refractive index change as a function of carrier density.

The physical mechanism responsible for the aforementioned improvements is actually more general than the Foupled cavity context under which it is presented here. It involves the introduction of any dispersive loss (or gain) mechanism into the laser, followed by deliberate detuning of the lasing frequency away from the loss (gain) peak (hence the name “detuned loading”). This more general picture, described in detail elsewhere [34] , predicts that the same improvements should also occur in semiconductor,lasers with distributed feedback.

7 ,

6 -

5 - 1.5

2.0

-5 -6

1-2.0 Fig. 15. Modulation corner frequency squared, normalized by their

conventional (no coupled cavity) values, plotted against the detuning parameter 6,

It is known that quantum well lasers, with their modified two- dimension density of states function, have a higher overall differential optical gain. It follows that modulation performance can be significantly improved in these quantum well lasers as compared to conventional devices. Results of theoretical calculations [35] of quantum’ well lasers are shown in Fig. 16, which shows the modulation bandwidth as a.function of well width L, in GaAs. The broken line gives the values for a conventional double heterostructure laser. As shown in the figure, it should be possible to double f , in a quantum well (2D) over its value in conventional (3D) lasers using L, of less than 80 a. If we further reduce the dimensionality of the electrons by confinhg it in a quantum wire structure whose dimensions are L, X L,, the differential gain will be further improved, as shown in Fig. 17, which shows the calculated values forf, as a function of L, (=i,). These results indicate that f, can be made about three times larger than that of a conventionallaser. Experimental work is under way to demonstrate the above theoretical predictions.

VIII. INTERMODULATION PRODUCTS One important application for a multi-gigahertz bandwidth

semiconductor laser is multichannel frequency division multi- plexed transmission of analog or microwave signals. An obvious quantity of concern here is the nonlinear distortion characteristics of’the laser. It is well known that a well-behaved semiconductor laser (i.e., those with a linear light-current characteristic without kinks and instabilities above lasing threshold) exhibits very little nonlinear distortion when modulated at low frequencies (below a few tens of megahertz) [36] . This is to be ‘expected since, at such low modulation speeds, the laser is virtually in a quasi-steady state as it is ramping up and down along the light-current curve, and consequently, the linearity of the modulation response is basically that of the CW light current characteristic, which is excellent in well-behaved laser diodes. Measurements and analysis have shown that second harmonic distortions of lower than -60 dB can be readily accomplished at the low frequency range [36] . However, it was also shown that as the modulation frequency increases, the harmonic distortions increase very rapidly-ai modulation frequencies above 1 GHz, the second and third harmonics can be as high as - 15 dB at a moderate optical modulation depth (-70 percent) [37] -[39] . These results can be well explained


I I h I I C

T = 3 0 0 K

0 1 ' ' " " I 50 I O 0

' 1.0 200

WELL WIDTH (8) Fig. 16. The CY parameter, a measurement of the intrinsic linewidth of a

semiconductor laser, and modulation corner frequency f, of a quantum well laser as a function of well width. The photon lifetime is taken to be 2.6 ps, and the photon density is 3.8 X 1013 cmW3.

25 U

3.0 ~

0 t-

3 I- z W

E 2.0 5

0

I Z W

I + 0

z ?I

-. 1.0

WELL WIDTH (8) Fig. 17. Same as for Fig. 16 but for a quantum wire laser.

by a perturbative analysis of the laser rate equations, which depicts the interaction of the photon and electron fluctuations as the origin of the large harmonic distortions observed at high frequencies.

In actual applications of multichannel signal transmission where baseband signals from different channels are carried on a number of well-separated high frequency carriers, second (or higher order) harmonic distortions generated by signals in a channel are actually of little concern since they generally do not fall within the frequency band of that particular channel (or, if the frequency bands are properly chosen, any other channels). The relevant quantity of concern here is the third- order intermodulation (IM) product of the laser transmitter: two signals at frequencies w1 and w2 within a certain channel can generate intermodulation products at frequencies 2wl - w 2 and 2 0 , - w1 , which under most circumstances lie within that particular channel and are thus objectionable. Questions of IM product dependence on modulation depth, signal frequencies, magnitude of relaxation oscillation, etc., need to be answered.

We have studied the IM characteristics of high-speed laser diodes capable of being modulated at multi-gigahertz frequencies. The experimental study consists of modulating the lasers with two sinusoidal signals, 20 MHz apart, and observing the various sum, difference, and harmonic frequencies thusgenerated. The

major distortion signals considered here are shown in Fig. 18. The principal distortion signals of practical concern, as mentioned above, are the third-order IM products, at frequencies 2 u l - w 2 and 20, - wl. The various distortion signals are systematically studied as one varies the frequency (a) of the modulating signals (the two signals are at w and w t 27r X 20 MHz), the optical modulation index (OMI), and the laser bias level. The OM1 is defined as B/A where B is half of the peak- to-peak amplitude of the modulated optical waveform and A is the average value. The major observed features are summarized as follows.

1) At low modulation frequencies (a few hundred megahertz) all the lasers tested exhibit very low IM products of below -60 dB (relative to the signal amplitude) even at an OM1 of approaching 100 percent.

2) The second harmonics of the modulation signals increase roughly as the square of the OMI, while the IM products increase as the cube of the OMI.

3) The relative amplitude of the IM product (relative to the signal amplitude) increases at a rate of 40 dB/decade as w increases, reaching a plateau at one half of the relaxation oscillation frequency, and resumes the 40 dB/decade increment as w exceeds the relaxation oscillation frequency. A typical value of the IM product at the plateau is -45 to -50 dB at an OM1 of 50 percent.

4) In some lasers the IM product may show a peak at one half of the relaxation oscillation frequency. The magnitude of this peak is found to roughly correspond to the magnitude of the relaxation oscillation resonance peak in the small signal modulation response of the laser.

Fig. 19 shows the IM and harmonic distortions of a high-frequency laser diode under the two-tone modulation as described above, at w = 27r X 3 GHz, at various OMI's. The relaxation oscillation frequency of this laser is at 5.5 GHz. The observed data of IM and second harmonic as functions of OM1 and fare plotted in the graphs in Figs. 20 and 21, respectively; the various curves in those graphs are from theoretical calculations [40]. The analytical results shown in Figs. 20 and 21 as solid curves are based on a small signal perturbative analysis of the rate equations, taking into account all eight frequency components shown in Fig. 18. The two solid curves in Fig. 18 are calculated results for two cases for which the relaxation resonance peak in the modulation response of the laser is or is not damped. The general trend of the experimental data agrees well with theoretical predictions.

The above results are significant in that 1) the linearity of the CW light-current characteristic (as well as distortion measurements at low frequencies) is not a reliable indication of its performance at high frequencies; 2) although the IM product initially increases at an alarming rate of 40 dB/decade as the modulation frequency is increased, it does settle to a steady value of --45 dB, which is satisfactory for many applications, including, for instance, television signal transmission.

IX. INTENSITY NOISE It has been recognized that noise fluctuations in lasers can be

divided into the categories of power fluctuation noise (or intensity fluctuation noise) and frequency or phase fluctuation


... ". modulatlan reswnse of loser

frequency

Fig. 18. Sidebands and harmonics generated by a two-tone modulation of a laser diode. This simulates narrow-band signal transmission at a high carrier frequency.

/ I -701 I I I I I I I I

0.1 0.4 0.6 0.8 1.0

Optical modulation depth(0MD) Fig. 20. Plots of second harmonic and third-order IM amplitudes as a

function of optical modulation index (OMI), with the signal frequency at 2 GHz. Experimental data points obtained with a high-frequency laser diode are also shown.

-20 -

-30 -

-40- U 0) -0

- .- - 2 -50-

E? a -60-

OMD : 80%

IM product [ 2 w , - w 2 , 2 w 2 - w , l

/ 0

Frequency (GHz)

Fig. 21. Plots of third-order IM amplitudes as a function of signal frequency, at an OM1 of 80 percent.

Fig. 19. Harmonics and IM products generated in a high-speed laser diode under a two-tone modulation. The two tones are 20 MHz apart centered at -3 GHz.

noise. Semiconductor lasers described in the previous sections are intended primarily for use in direct detection transmission systems (rather than coherent transmission systems), and consequently, the intensity fluctuation power spectrum is of primary concern. The theoretical intensity fluctuation noise spectrum can be calculated using the simple rate equations including the Langevin force terms. The result is [41]

(2-V +a where p o is the steady-state photon density, wR is the relaxa-

tion oscillation frequency, 1/rR = 1/r, + A p o , rs is the spontaneous lifetime, and A is the differential optical gain, identical to that in (2). The denominator in (10) is identical to that in the modulation response of injection lasers and is responsible for the second-order low-pass network nature of the response. An important quantity of practical interest is the relative intensity noise (RIN), which is defined as follows:

where (LIP)' is the mean square intensity fluctuation spectral density of the light output, and Po is the average light output power. This quantity expresses the maximum achievable signal- to-noise ( S / N ) in a laser transmission system where the dominant noise source is the laser noise (which is generally the case for short or medium length links where the transmission loss is low). Assuming that the laser is biased a certain dc power level

,132 IEEE JOURNAL O F QUANTUM ELECTRONICS, VOL. QE-21, NO. 2, FEBRUARY 1985

i 18mA

iphotodiode= 148tLA

= 2 5 m A = 321 p A

= 3 4 m A = 5 1 6 p A

0 l i gh t

RF amplifier gain = 2 0 d B Frequency range = 1.8 - 4.1 G H z

RF amplif ier gain = 2 0 d B Frequency range = 1.8 - 4.1 G H z

Fig. 22. Intensity noise of a window BH on SI laser as observed at the output of a p-i-n photodiode, in the frequency range of 1.8 and 4.1 GHz. The dc photocurrents at each bias level are also shown. The RF output from the p-i-n diode is am- plified by 20 dB. The noise background of the amplifier is shown in (a) with the laser light blocked. Notice that the reference level (the RF power level at the top of the display) is different from the rest.

and modulated by a sinusoidal signal with 100 percent modulation depth, the S/N will be (RIN - 3 dB)/unit frequency. If the modulation depth is less than 100 percent, the S/N should be reduced according to the reduction of the signal strength. In view of the resonant nature of the noise spectrum density [see (lo)] , the maximum achievable S/N depends on the signal frequency, and it is obvious that for a fixed signal amplitude the S/Nwill be lowest near the relaxation oscillation resonance frequency. Using (10) and for w >> l/rR (whch is typically less than a few hundred megahertz), the RIN takes the following form:

(12)

The form of the RIN as given in (12) shows that high-speed lasers with a high photon density p o in the active region, such as the window laser described in Section V, are much superior to conventional lasers as transmitter.s, because 1) the relaxation oscillation resonance can be pushed to a very high frequency far away from the frequency band of interest, thus avoiding the excess noise near the resonance, and 2) the increased optical power density brings about a lower overall RIN according to the l/p: relation.

In an actual experiment the RIN is measured by detecting the light output from a laser using a photodiode, whose electrical output, after amplification, is displayed on a microwave spectrum analyzer. The dc photocurrent ( z p d ) is also recorded simultaneously. The spectrum analyzer display corresponds to the ( A P ) 2 in (11) with the value of (Po)’ represented by 50 R X I&. Here, caution must be taken since most photodiodes (especially avalanche photodiodes) show an excess dc gain of several decibels over the high frequency response, and therefore, the value of RIN obtained in this manner may be lower than the actual value. Actual measurement data taken on a window BH on SI laser are shown in Fig. 22(a)-(d). One clearly sees the relaxation oscillation resonance “sweeping” through the frequency band displayed in the figure as the op-

tical power is increased. A 20 dB gain amplifier follows the p-i-n detector used in this measurement in obtaining the displays in Fig. 22. The background noise level, with the amplifier in place, is shown in Fig. 22(a). (Notice that the reference level, which is the RF power level at the top of the display, in Fig. 22(a) is different from the rest of the displays.) The RIN of the laser shown in Fig. 22 is plotted in Fig. 23. In the 2-4 GHz range, the RIN lies between - 150 and - 145 dB/Hz, low enough for many low-noise wide-band applications requiring a high dynamic range.

X. SUPERLUMINESCENT DAMPING The variation of the magnitude of the relaxation oscillation

resonance among laser structures has been attributed to a variety of causes, including the magnitude of the spontaneous emission factor and lateral carrier diffusion. The existence of this resonance peak limits the useful modulation bandwidth of lasers to considerably below the relaxation oscillation frequency, where the response is relatively flat. Although equaliza- tion by external electrical circuit elements [42], [43] or external light injection [44] , [45] have been suggested as means of suppressing the resonance, it is far preferable that the laser possesses an intrinsically weak resonance. Lateral carrier diffusion, which plays a dominant role in some laser structures, may not be an effective means for suppressing relaxation oscillation, especially when the resonance occurs at high frequencies. The reason is that the effect of the diffusion is that of a low- pass filter on the modulation current [46], 1471 , thus creating a dip in the modulation response before the onset of the resonance peak, and it does not actually remove the resonance [48] ,

There is another mechanism that can lead to a flat frequency response. This mechanism involves the nonuniform distribu- tion of photons and carriers along the length of a laser cavity with a small mirror reflectivity. This effect is particularly useful in practice since the facet reflectivity can be modified quite easily.

To analyze this effect, the spatially uniform rate equations commonly employed are not valid and one must use the local

[491.


-120, I Shaded reglans represents sensltlvlly lirnbt of detectton system I

z -130

.- 0 Z

l0.5rnW I W -150 N,,,,, N,,

1.8 Frequency (GHz)

4.1

Fig. 23. The relative intensity noise (RIN) of the laser shown in Fig. 26.

photon and injected carrier conservation equations [50] . Fig. 24 shows the numerically calculated amplitude modulation response of five lasers [51]. One of the lasers corresponds to a common laser with a facet feedback of 0.3 from both facets and a length of 300 pm. In each of the other diodes, one of the facets has a reflectivity of 0.3, whereas the other facet takes on reflectivities of 0.1, 0.01 ,0.005, and 0.0001. In each of these four cases, the diode length is adjusted so that they possess a photon lifetime identical to that of the first laser diode. All five lasers possess the same threshold gain of 100 cm-' . The devices are pumped to an identical level corresponding to an unsaturated gain of 200 cm-' . The commonly employed rate equations analysis predicts that these five diodes have identical output powers and frequency responses, in obvious disagreement with that shown in Fig. 24. Notice that for the two cases where R 1 = 0.3 and 0.1 , the amplitude curves are almost identical to that calculated from the simple rate equations, displaying a resonance at 2.9 GHz. The resonance is suppressed in devices with a smaller feedback from a mirror and at a value of R1 between 0.001 and 0.005, the response is maximally flat. Further decrease in R reduces the modulation bandwidth. These data show that the simple rate equations fail to give the correct shape of the frequency response when the reflectivity of one of the mirrors is reduced to below 0.1, but does, however, give the correct comer frequency even when the reflectivity of one of the mirrors is as low as lom3. These conclusions are not restricted to the particular case above but are, in general, true over a wide range of pump currents and photon lifetimes.

The above theoretical predictions are confirmed experimentally [52] using high-speed BH on SI lasers. Antireflection (AR) coatings were evaporated onto both mirror facets. The reflectivity was estimated from the increase in lasing threshold to be between 1 and 5 percent. Accompanying the increase in lasing threshold are the softening of the lasing transition and the increase in the number of lasing modes. The multimode lasing spectrum in AR coated lasers has been observed before [53] , [54] and was explained by a theory involving effects of superluminescence inside the laser cavity [55] . An immediate con- sequence of the increase in the number of lasing modes is the increase in the spontaneous emission factor, which has the

FREQUENCY(GHz) Fig. 24. Amplitude responses of diodes with different mirror reflec-

tivities but different lengths (so that they all possess the same photon lifetimes of 1 ps). They are pumped to an unsaturated gain of 200 cm-l, which is above the classical threshold of all of these lasers.

effect of reducing the magnitude of the relaxation oscillation resonance. This, in conjunction with additional effects due to superluminescence as described before, results in a modulation response which is completely flat up to multi-gigahertz frequencies.

The high frequency modulation results are shown in Fig. 25, where the different curves are obtained with the laser biased at different power levels. A totally flat response with a 3 dB bandwidth of 5 GHz was obtained with this laser. The flatness of these response curves was in marked contrast to those commonly observed in regular injection lasers without coatings.

XI. LONGITUDINAL MODE SPECTRUM In short-haul applications (<I km) where multimode optical

fibers are the optimal choice, the longitudinal mode spectrum is not a cause for concern since the primary dispersion of multimode fibers is intermodal instead of chromatic. For applications in a long transmission link (several kilometers) where single-mode fibers are used, the spectrum of the laser becomes important. Ideally, one would prefer that the laser transmitter in such a system oscillates in a single longitudinal mode even under high-speed modulation. In reality, no laser oscillates in a true single mode, and the degree of suppression of non- lasing modes which can be tolerated depends on the application. In medium distant (<lo km), repeaterless single-mode fiber transmission, a sappression ratio of 30-50 appears to be sufficient. In some specialized very-long-haul transmission (tens to over 100 km) such as the transatlantic cable system, suppression ratios in the thousands are required. The latter is not attainable by conventional lasers, and specialized structures such as the C 3 (cleaved coupled cavity) [56] or distributed feedback lasers must be used. Most well-behaved, conventional index guided lasers can satisfy the 30-50 suppression ratio re- quirement for medium distance transmission under CW operation. It is well known that such lasers may not maintain single- mode operation under high-speed modulation. It is important to understand and delineate the drive conditions under which the laser can preserve its essentially single-mode spectrum.

Previous experiments have shown that when microwave modulation is applied to a single-mode laser, the lasing spectrum will remain single-mode unless the optical modulation


Frequency

Fig. 25. Modulation response of an AR coated laser at various bias CUT- rents. The length of the laser is 300 pm, and the laser is biased at (a) 0.3, (b) 0.6, (c) 1, (d) 3, and (e) 5 mW, respectively.

depth exceeds a certain critical level [57] . This section ad- dresses these issues and provides an analytical treatment of the phenomenon.

In our experiments, we investigate the time-averaged lasing spectrum of lasers under high-frequency continuous microwave modulation at various frequencies and modulation depths. The lasers used are index guided (BH on SI) lasers with a stable single transverse mode. The longitudinal mode spectrum becomes essentially single-moded at an output power slightly below 1 mW and -1.3 mW for a 120 pm and a 250 pm laser, respectively. The fraction of power contained in the dominant lasing mode is higher in the short laser than the long one at all corresponding output power levels. All the lasers tested retain their single-mode spectrum unless the optical modulation depth, defined as the ratio of the amplitude to the peak of modulated optical waveform, exceeds a critical level at -75-90 percent, depending on the purity of the original CW lasing spectrum. An interesting observation is that, contrary to common belief, this critical modulation depth does not depend on modulation frequency, at the frequency range above -5OOMHz. However, it can be observed that the width of the individual modes broadens at higher frequencies, although the relative amplitudes of the modes do not change. This arises from fluctuations in the refractive index of the cavity as a result of fluctuation in carrier density. A simple single-mode rate equation Analysis shows that under a constant optical modulation depth, the fluctuation in carrier density increases with increasing modulation frequency, and consequently, the line broadening effect is more visible at high frequencies. A rough estimate of the time constant involved in spectral

transients can be obtained by considering the time evolution equation for the fraction of power in the dominant mode (a,) and let /3 + 0, in which case one find that the characteristic time constant is given by [58]

r-z 2 f l P (1 3)

where b is the width of the Lorenzian gain lineshape as a function of longitudinal mode number [58], and T~ is the photon lifetime. Since b - and T~ - 2 ps, T is of the order of 0.5 ns, which agrees well with experimental observations [59] and

numerical results [60] for GaAs lasers. Hence, if one modulates a laser at above -500 MHz, the spectrum does not ‘have sufficient time to respond within a modulation cycle, and consequently, the spectral content will not vary significantly with time. A simple analysis based on averaging of the time evolution equation of fractional modal power shows that when a laser is biased at a certain optical power and being modulated at high frequencies with an optical modulation depth of r), the time-averaged lasing spectrum is equivalent to that of the laser operating CW (without modulation) at a reduced power level of SA given by

where So is the actual average output power from the laser. [The optical modulation depth q defined here is the ratio of the amplitude to the peak of the optical waveform, i.e., if the optical waveform is So t S1 cos w t then 17 = 2S1 /(SI t So).] Fig. 26 shows a plot of the apparent reduction factor S/,/S, versus q. The results shows that high-frequency modulation has little effect on the lasing spectrum unless the optical modulation depth exceeds -80 percent. The points shown on the same plot are experimental results.

If one further increases the microwave drive to the laser beyond the point of 100 percent optical modulation depth (So = S,), the bottom of the optical waveform will clip. The integral ( l / S ( t ) ) becomes very large and consequently SA becomes very small. The spectrum would look like that of a laser below lasing threshold. This is in accord with experimental observations.

The analysis presented above is based on a strictly homogen- ously broadened gain system and, therefore, does not take into account mode jumping and spectral gain suppression. Spectral gain suppression is manifested as a decrease in the actual amplitude of the nondominating longitudinal modes as the total optical power is increased, and is usually observed only at fairly high optical power levels. This phenomenon has been explained by considering nonlinear optical properties of the semiconductor material [61] and actually aids the laser in maintaining a single-mode spectrum under high-frequency mqdulation.

VII. LARGE SIGNAL EFFECTS As noted in Section I1 of this paper, the bulk of the experi-

mental and theoretical results presented in the previous sections are concerned with small signal modulation. By “small signal” one means that the modulation depth in the optical output of the laser is substantially less than 1. In wide-band digital,telecommunication, the on-off ratio in the laser output should be as large as possible to avoid excessive power penalty. The modulation response of a laser under large modulation is different from (and generally worse than) the small signal response, although the latter is often a good indicator of the former as long as the laser is maintained above threshold at all times. Ikegami and Suematsu [62] studied large signal effects by direct numerical integration of the rate equations: The result is quite similar to the small signal response except that the relaxation oscillation occurs at a lower frequency. The reason for this is that the laser cannot be swiftly brought from a state


r I

0 u 0.2 0.4 0.6 0.8 1.0

17 Fig. 26. A plot of Sb/So versus 11. Sb is the photon density correspond-

ing to a bias level, at which the laser would emit a longitudinal mode spectrum similar to that when the actual bias level is So and the laser is modulated at high frequencies. The vertical bars are derived from experimental observations under modulation at 1-4 GHz.

input current drive

laser optical output Fig. 27. Input current pulses (top row) and responses of a window BH on SI laser (bottom row) at 70 percent optical mod-

ulation depth. The pulses are negative-going and are inverted in these displays.

of very low optical ievel to a high level, due to the low rate of stimulated emission at the low-power state. This can be illustrated by modulating a laser, which is biased at a high optical level, with two closely spaced negative-going current pulses. The result is shown in Fig. 27, where the top row of pictures show the current pulses at various spacings, and the bottom pictures show the corresponding optical output from a window BH on SI laser biased at 10 mW. The optical modulation depth in Fig. 27 is 70 percent (i.e., the optical power at the bottom of the pulses is 30 percent of the bias level), and it is clear that the optical output is a faithful reproduction of the current pulse input up to a pulse spacing of 100 ps, which is limited by the current pulse width. The effect of increasing the optical modulation depth is shown in Fig. 28. When the optical modulation depth exceeds -85 percent, the laser responds slug- gishly when going from the “low” to the “high” state, and severe relaxation oscillation occurs. This is obviously not satisfactory for a pulse-code modulation transmission. The slew rate of the optical output in the large signal regime has been calculated from the rate equations by Kaminow and Tucker [63] , [64] , and the relaxation oscillation associated with large

modulation depth has recently been observed by Eisenstein et al. [65] . The basic conclusion of this section is that when high-speed modulation is desired, the optical modulation depth should be maintained below -70 percent, but when power penalty must be minimized in a digital telecommunication system, modulation speed will be sacrificed.

XIII. CONCLUSION AND OUTLOOK The advances in high-speed semiconductor laser technology

described in this paper have opened up new areas of applications involving high-data-rate, high-speed analog and microwave transmission through optical fibers. Digital transmission in tens of gigabits/second, attractive as it may see~m, may not be practically feasible in the near future, at least not on a large scale, due to the still immature state of gigabit electronic technology. But in areas of high-speed analog or microwave transmission where coaxial cables and metallic waveguides have’ been the only available transmission channels, and where optical transmission techniques have never been considered viable, the new developments in high-speed lasers have made a significant impact. High-speed semiconductor lasers with direct


dr ive currenl

optlcal response

d r l ve current

optical O U i D U t

Fig. 28. Effect of increasing optical modulation depth on pulse modulation performance, illustrated by a double negative-pulse experiment for two pulse spacings. The displays are inverted, and the arrows on the left-hand side indicate the zero optical level.

modulation bandwidths in the upper gigahertz range are now commercial realities, and are being used in systems ranging from radar signal transmission to locking of oscillators from remote stations. A very interesting area of application of these high-speed lasers under consideration involves optically inter- connecting electronic chips within computers. The high-speed nature of these laser devices, together with concurrent developments in integrated optoelectronic technology [65] , make this idea ever more attractive. As noted earlier in this pkper, following the trend of recent developments, lasers with direct modulation bandwidths beyond 10 GHz should become a reality in the not-too-distant future. Thiswill inspire even more

flew applications of optical techniques in the area of high-speed signal transmission and processing.

ACKNOWLEDGMENT

Most of the work described in this paper was done in col- laboration with Dr. N. Bar-Chaim, Dr. I. Ury, and Dr. C. Harder. Many enlightening discussions with Dr. S. Margalit have been most useful.

REFERENCES

[ l ] H. Statz and G. deMars, “Transients and oscillation pulses in masers,” in Quantum Electronics, C. H. Towns, Ed. New York: Columbia Univ. Press, 1960, pp. 530-537.

1

1

L. Figueroa, C. Slayman, and-H. W. Yen, “High frequency characteristics of GaAlAsinjection lasers,”IEEEJ. Quantum Electron.,

H. Kressel and J , K. Butler, Semiconductor Lasers and Hetero- junction LEDs. New York: Academic, 1979. T.Ikegami and Y. Suematsu, “Direct modulation of semiconductor junction lasers,” Electron. Commun. Japan, vol. B51, pp. 51-58, 1968. T. P. Paoli and J . E. Ripper, “Directmodulationof semiconductor lasers,”Puoc. IEEE, vol. 58, pp. 1457-1465, 1970.

V O ~ . QE-18, pp. 1718-1727, 1982.

[6] S. I . Gonda and S. Mukai, “Degradation and intensity fluctuation in CW GaAlAs double heterostructure lasers,” IEEE J. Quantum Electron. ,vol. QE-11, pp. 545-550, 1975.

[7] J . Angerstein and D. Siemsen, “Modulation characteristics of injection lasers including spontaneous emission-2. Experiments,” Arch. Elek. Ubertragung., vol. 30, pp. 477-480, 1976.

[8] T. P. Paoli, “Magnitude of the intrinsic resonant frequency in a semiconductor laser,” Appl. Phys. Lett., vol. 39, pp. 522-524, 1981.

[9] F. Stern, “Calculated spectral dependence of gain in excited GaAs,”J. Appl. Phys.,vol. 47, pp. 5382-5386, 1976.

[ l o ] K. W. Wakao, N. Takagi, K. Shima, K. Hanamitsu, K. Hori, and M. Takusagawa, “Catastrophic degradation level of visible and infrared GaAlAs lasers,” Appl. Phys. Lett., vol. 41, pp. 1113- 1115, 1982.

[ l l ] A “crank” TJS laser can operate reliably at a pump current density equivalent to almost 10 kA/cm2 without suffering excessive heating or undue degradation. Data from Mitsubishi Corp.

[ 121 N. Bar-Chaim, J . Katz, I. Ury, and A. Yariv, “Buried heterostructure AlGaAs laser on semi-insulating substrate,” Electron. Lett.,

[ 131 I. Ury, K. Y. Lau, N. BarChaim, and A. Yariv, “Very high frequency GaAlAs laser-field effect transistor monolithic integrated circuit,” Appl. Phys. Lett., vol. 41, pp. 126-128, 1982.

[14] N. BarChaim, K. Y. Lau, I. Ury, and A. Yariv, “High speed GaAlAs/GaAs pin photodiode on a semi-insulating substrate,” Appl. Phys. Lett., vol. 43, pp. 261-262,1983.

[ 151 H. C. Casey and M. B. Panish, Heterostructurehsers. New York: Academic, 1978, part A, p. 174.

[16] H. Kressel and J. K. Butler, Semiconductor Lasers and Hetero- junction LED’s. New York: Academic, 1975, p. 230.

[I71 S. Takahashi, T. Kobayashi, H. Saito, and Y. Furukawa, “GaAs- AlGaAs DH lasers with buried facets,” Japan J. Appl. Phys., vol.

[ 181 H. Blauvelt, S Margalit, and A. Yariv, “Large optical cavity AlGaAs buried heterostructure window laser,”AppZ. Phys. Lett.,

[19] K. Y. Lau, N. BarChaim, I. Ury, and A. Yariv, “An 11 GHz direct modulation bandwidth GaAlAs window laser on semi-insulating

45, Aug. 1984. substrate operating at room temperature,” Appl. Phys. Lett., vol.

[20] K. Y. Lau and A. Yariv, “Effect of superluminescence on the modulation response of semiconductor lasers,” Appl. Phys. Left., vol. 40, pp. 452-454, 1982.

[21] K . Y. Lau, N. BarChaim, I. Ury, C. Harder, and A. Yariv, “Super- luminescent damping of relaxation resonance in the modulation response of GaAs lasers,” Appl. Phys. Lett.,vol. 43, pp. 329-331, 1983.

V O ~ . 17, pp. 108-109, 1981.

17, pp. 865-870,1978.

VOI. 4 0 , ~ ~ . 1029-1031,1982.


[321

341

J. Helszajn, Passive and Active Microwave Circuits. New York: Wiley, 1978. Appl. Note on s-parameters, Hewlett Packard Corp. J. Katz, S. Margalit, C. Harder, D. P. Wilt, and A. Yariv, “The intrinsic equivalent circuit of a laser diode,” IEEE J. Quantum Electron., vol. QE-17, pp. 4-7, 1981. J. D. Jackson, Classical Electrodynamics, 2nd ed. New York: Wiley, 1975. W. Harth and D. Siemsen, “Modulation characteristics of injection lasers including spontaneous emission-1. Theory,” Arch. Elek. Ubertragung., vol. 30, pp. 343-348, 1976. K. Aiki, M. Nakamura, T. Kuroda, K. Umeda, R. Ito, N. Chinone, and M. Maeda, “Transverse mode stabilized AlGaAs injection laser with channeled substrate planar structure,” IEEE J. Quantum Electron., vol. QE-14, pp. 89-94, 1978. K. Furuya, Y. Suematsu, and T. Hong, “Reduction of resonance- like peak in direct modulation due to carrier diffusion in injection lasers,” Appl. Opt., vol. 17, pp. 1949-1952, 1978. M. Maeda, K. Nagano, M. Tanaka, and K. Chiba, “Buried heterostructure laser packaging for wideband optical transmission system,” IEEE Dans. Commun., vol. COM-26, pp. 1076-1081, 1978. L. Figueroa, C. Slayman, and H. W. Yen, “High frequency characteristics of GaAlAs injectionlasers,”IEEEJ. Quantum Electron.,

R . S . Tucker and D. J . Pope, “Microwave circuit model of semiconductor injection laser,” IEEE J. Microwave Theory Tech.,

C. P. Lee, S . Margalit, L. Ury, and A. Yariv, “GaAsCaAlAs injection lasers on semi-insulating substrate using laterally diffused junctions,”Appl. Phys. Lett.,vol. 32, pp. 410-412, 1978. N. Bar-Chaim, J . Katz, I. Ury, and A. Yariv, “Buried heterostructure AlGaAs laser on semi-insulating substrate,” Electron. Lett.,

K. Vahala and A. Yariv, “Detuned loading in coupled cavity semiconductor lasers-Effect on quantum noise and dynamics,” Appl. Phys. Lett., to be published. Y. Arakawa, K. Vahala, and A. Yariv, “Quantum noise and dynamics in quantum well lasers,’’ Appl. Phys. Lett., to be published. K. E. Stubkjr, “Nonlinearity of D. H. GaAlAs lasers,” Electron. Lett.,vol. 15,pp. 61-63, 1979. T. Ikegami and Y. Suematsu, “Direct modulation of semiconductor junction lasers,” Electron. Commun. Japan, vol. 53B, pp. 69- 75, 1970. T. H. Hong and Y. Suematsu, “Harmonic distortions in direct modulation of injection lasers,” Pans. IECE Japan, vol. E62, pp.

K. Y. Lau and A. Yariv, “Non-linear distortions in the current modulation of non-self-pulsing and weakly self-pulsing GaAs/ GaAlAs injection lasers,” Opt. Commun., vol. 34, pp. 424-428, 1980. - , “Intermodulation distortion in directly modulated semiconductor laser,”Appl. Phys. Lett., vol. 45, pp. 1034-1037, 1984. K. Vahala and A. Yariv, “Semiclassical theory of noise in semiconductor lasers-Part 11,” IEEE J. Quantum Electron., vol. QE-

Y. Suematsu and T. Hong, “Suppression of relaxation oscillation in the light output of injection lasers by electrical resonance circuit,” IEEE J. Quantum Electron., vol. QE-13, pp. 756-762, 1977.

V O ~ . QE-18, pp. 1718-1727, 1982.

V O ~ . MTT-31, pp. 289-294,1983.

V O ~ . 17, pp. 108-109, 1981.

142-146,1979.

19,pp. 1102-1109,1983.

[43] J. Katz, S. Margalit, and A. Yariv, “A proposed new method for damping relaxation oscillation in laser diodes,” IEEE Electron. Devices Lett., vol. EDL-3, pp.,333-335, 1982.

[44] R. Lang and K. Kobayashi, “Suppression of the relaxation oscillation in the modulated output of semiconductor lasers,”IEEE J. Quantum Electron., vol. QE-12, pp 194-199, 1976.

[45] D. Fekete, W. Streifer, D. R. Scrifes, and R. D. Burnham, “High speed laser modulation with integrated optical injection,” Appl. Phys. Lett., vol. 37, pp. 975-977, 1980.

[46] N. Chinone, K. Aiki, M. Nakamura, and R. Ito, “Effect of lateral mode and carrier density profile on dynamic behavior of semiconductor lasers,” IEEE J. Quantum Electron., vol. QE-14, pp.

[47] D. P. Wilt, K. Y. Lau, and A. Yariv, “The effect of lateral carrier diffusion on the modulation response of a semiconductor laser,”

625-631, 1978.

J. Appl. PhyS.,vol. 52, pp. 4790-4794, 1981.

[48] K. Furuya, Y. Suematsu, and T. Hong, “Reduction of resonance- peak in direct modulation due to carrier diffusion in injection lasers,”Appl. Opt., vol. 17, pp. 1949-1952, 1978.

[49] L. Figueroa, C. Slayman, and H. W. Yen, “High frequency characteristics of GaAlAs injection lasers,” IEEE J. Quantum Elec- tron.,vol. QE-18, pp. 1718-1727, 1982.

[50] W. E. Lamb, “Theory of an optical maser,’’ Phys. Rev. ,vol. 134A, pp. 1429-1450, 1964; see also, A. Icsevgi and W. E. Lamb, “Propagation of light pulses in a laser amplifier,” Phys. Rev.,

[51] K. Y. Lau and A. Yariv, “Effect of superluminescence on the modulation response of semiconductor lasers,” Appl. Phys. Lett., vol. 40, pp. 452-454, 1982.

[52] K. Y. Lau,I. Ury, N. Bar-Chaim, C. Harder, and A. Yariv, “Super- luminescent damping of relaxation resonance in the modulation response of GaAs lasers,” Appl. Phys. Lett., vol. 43, pp. 329- 331,1983.

[53] R. G. Plumb and J . P. Curtis, “Channel substrate narrow stripe GaAslGaAlAs lasers with quarter wavelength facet coatings,” Electron. Lett., vol. 16, pp. 706-707, 1980.

[54] M. Ettenberg, D. Botez, D. B. Gilbert, J . C. Connelly, and H. V. Kowger, “The effect of facet mirror reflectivity on the spectrum of single mode CW constricted double heterojunction diode lasers,” IEEE J. Quantum Electron., vol. QE-17,pp. 2211-2214, 1981.

[55] K. Y. Lau and A. Yariv, “A theory of longitudinal modes in semiconductor lasers,” Appl. f iys. Lett . , vol. 42, pp. 763-765, 1983.

[56] W. T. Tsang, N. A. Olsson, and R. A. Logan, “High speed direct single-frequency modulation with large tuning rate and frequency excursion in cleaved coupled cavity semiconductor lasers,” Appl. Phys. Lett., vol. 42, pp. 650-652, 1983.

[57] M. Nakamura, K. Aiki, N. Chinone, R. Ito, and J. Umeda, “Longi- tudinal mode behavior of mode-stabilized AlGaAs injection lasers,” J. Appl. Phys, vol. 49, pp. 4644-4648, 1978.

[58] K. Y. Lau, C. Harder, and A. Yariv, “Longitudinal mode spectrum of semiconductor lasers under high speed modulation,” IEEE J. Quantum Electron., vol. QE-20, pp. 71-79,1984.

[59] F. Mengel and V. Ostoich, “Dynamics of longitudinal and transverse modes along the junction plane in GaAlAs stripe lasers,” IEEEJ. Quantum Electron., vol. QE-13, pp. 359-361,1977.

[60] K. Petermann, “Theoretical analysis of spectral modulation behavior of semiconductor injection lasers,’’ Opt. ,Quantum Elec- tron.,vol. 10, pp. 233-242, 1978.

[61] M. Yamada and Y. Suematsu, “Analysis of gain suppression in undoped injection lasers,” J. Appl. Phys., vol. 52, pp. 2653-2664, 1981.

[62] I. P. Kaminow, “High speed modulation of semiconductor lasers,” presented at Conf. Opt. Fiber Commun., New Orleans, LA, 1984, paper MJ1.

1631 R. S . Tucker and I. P. Kaminow, “High speed switching transients in directly modulated InGaAsP lasers,” presented at CLEO ’84, Anaheim, CA, 1984, paper FF1.

[64] G. Eisenstein, U. Koren, R. S . Tucker, B. L. Kasper,A. H. Gnauck, and P. K. Tien, “High speed analog and digital modulation of 1.51 pm wavelength three channel buried crescent InGaAsP lasers,”Appl. Phys. Lett.,vol. 45, pp. 311-313, 1984.

[65] N. Bar-Chaim, I. Ury, and A. Yariv, “Integrated optoelectronics,” IEEE Spectrum, pp. 38-45, May 1982.

V O ~ . 185A,pp. 517-545,1969.

Kam Y. Lau (S’79-M’81) was born in October 1956. He received the B.S. and M.S. degrees in 1978 and the Ph.D. degree in electrical engineering in 1981, all from the California Institute of Technology, Pasadena.

He worked at Bell Laboratories, Holmdel, NJ, during the summers of 1977 and 1978. From 1979 to 1981 he worked at the Jet Propulsion Laboratory, Pasadena, on development of fiber- optic systems. While a t Cal Tech his primary research work was in high-speed semiconductor


lasers. He was a Research Fellow at Cal Tech from June to October, 198 1 , when he joined Ortel Corporation, Alhambra, CA, where he is con- tinuing work in semiconductor optoelectronics and related technologies.

Dr. Lau is a member of the Optical Society of America and the IEEE Quantum Electronics and Applications Society.

Amnon Yariv (S’56-M’59-F’70) was born in Tel Aviv, Israel, on April 13, 1930. He received the B.S., MS., and Ph.D. degrees in electrical engineering from the University of California, Berkeley, in 1954, 1956,and 1958,respectively.

A veteran of the Israeli war of independence, he came to the United States in 195 1. He went to Bell Laboratories, Murray Hill, NJ, in 1959, joining the early stages of the laser effort. He came to the California Institute of Technology, Pasadena, in 1964 as an Associate Professor of

Electrical Engineering, becoming a Professor in 1966. In 1980 he be- came the Thomas G. Myers Professor of Electrical Engineering and Ap- plied Physics. On the technical side, he took part (with various co-

workers) in the discovery of a number of early solid-state laser systems, in the formulation of the theory of parametric quantum noise and the prediction of parametric fluorescence, in the invention of the technique of mode-locked ultrashort-pulse lasers and FM lasers, in the introduction of GaAs and CdTe as infrared electrooptic and window materials, in proposing and demonstrating semiconductor-based integrated optics technology, and in pioneering the field of phase congugate optics. His present research efforts are in the areas of nonlinear optics, recombination mechanisms in semiconductors, semiconductor lasers and integrated optics, especially the problem of monolithic integration of transistor?, injection lasers, and aetectors for high-frequency applications. He authored or coauthorea some 300 papers in professional Journals, as well as a number of basic texts in quantum electronics, optics, and quantum mechanics. He is a Consultant to the Hughes Aircraft Research Laboratories, and is a founder and Chairman of the Board of Ortel Cor- poration, Alhambra, CA.

Dr. Yariv is a member of the American Physical Society, Phi Beta Kappa, and the National Academy of Engineering, and a Fellow of the Optical Society of America. He was the recipient of the 1980 Quantum Electronics Award of the IEEE. He is also an Associate Editor of Optics Communications and the Journal of Applied Physics, and was previously Associate Editor of the IEEE J O u R K . 4 L OF QIJANTUM Et.Ec’rwNIcs.

A Low Dark-Current, Planar InGa Photodiode with a Quaternary I

Cap Layer

A b s r r o ~ t - I n ~ . ~ ~ G a ~ , 4 ~ A s p-i-n photodiodes have become the most suitable photodectectors for long wavelength (1-1.65 pm) optical fiber communication systems due to their low dark-currents. Further reduction of the darkcurrent to subnanoampere range will bring significant improvement in receiver sensitivity at low bit rates. In this paper, we report the successful fabrication of I ~ o . ~ ~ G ~ ~ , ~ ~ A s p-i-n diodes with room temperature dark-currents as low as 0.1 nA at -10 V bias by in- troducing a quaternary InGaAsP cap layer on the ternary InGaAs. In this new structure the InGaAsP cap layer undergoes the surface treatment and faces the dielectric silicon nitride film, but the p-n junction is still located inside the InGaAs layer.

Since the addition of the wide bandgap cap layer results in low darkcurrent, the reduction of darkcurrent may be attributed to surface effects. Analysis of the temperature and voltage dependence of the darkcurrent indicates that ar room temperature the ohmic conduction through the shunt formed on the surface limits the darkcurrent in this new diode structure.

Manuscript received May 2, 1984; revised October 5, 1984. 0. K. Kim is with Collins Transmission Systems Division, Rockwell

B. V. Dutt is with AT&T Bell Laboratories, Holmdel, NJ 07733. R. J. McCoy and J. F.. Zuber are withAT&TBell Laboratories,Murray

International Corporation, Dallas, TX 75207.

Hill. NJ 07974.

L I. INTRODUCTION

IGHTWAVE communication systems in the long wavelength range of 1-1.65 pm have stimulated the develop-

ment of photodiodes fabricated from Inp,,3Ga0,7As lattice matched to InP substrates due to its bandgap of 0.75 eV. To date, a receiver using an InGaAs p-i-n diode and a GaAs field- effect transistor has exhibited high sensitivity [ I ] in comparison to other alternatives, and thus, the InGaAs p-i-n diode has become one of the more attractive photodetectors at the long wavelengths.

Recently, a planar type InGaAs diode has been demonstrated [2] , [3] to be comparable to a mesa configuration [4] - [6] in dark-current, quantum efficiency, and speed of photoresponse. Furthermore, the planar diode has proved to be more reliable and immune to harsh environment [7], [ 8 ] , when a SiN, dielectric film is utilized as a protection layer. The development of technology for manufacture of planar InGaAs photodiodes is therefore of current interest.

In this paper, we will discuss a new type of planar, back- illuminated InGaAs diode which exhibits dark-currents as low

Documents

IEEE JOURNAL OF QUANTUM ELECTRONICS, VOL. QE-21, NO. …klau/pubs/03.pdfIEEE JOURNAL OF QUANTUM ELECTRONICS, VOL.QE-21, NO. 2, FEBRUARY 1985 121 Papers Ultra-High Speed Semiconductor