1. Introduction
Driven by the rapidly increasing data capacity in optical networks and data centers, global network bandwidth demand is surging. A critical trend in fiber optic communication networks is the urgent need for high-speed, high-capacity systems. Directly modulated lasers (DMLs) have emerged as a compelling solution for passive optical networks (PON) and Ethernet applications, owing to their cost-effectiveness and low power consumption [
1,
2,
3]. Consequently, enhancing the modulation bandwidth of DMLs and simplifying their fabrication is paramount.
The two primary methods for enhancing the bandwidth of DMLs are increasing the relaxation oscillation frequency
and employing optical feedback techniques such as detuned loading (DL) and photon–photon resonance (PPR) effects [
4,
5,
6].
can be expressed as:
where
is the group velocity of light in the medium;
is the differential gain of the medium, where
g is the gain of the medium and
N is the carrier density in the medium;
is the internal quantum efficiency;
is the optical confinement factor;
e is the electron charge;
V is the volume of the active region;
I is the injection current of the laser; and
is the threshold current. Methods to enhance
include increasing the differential gain, reducing the threshold current, and decreasing the active region volume. However, due to damping effects, the maximum bandwidth is limited to
(where
K is the damping factor) [
7]. Researchers have explored DL and PPR effects to overcome this limitation [
8,
9,
10,
11,
12,
13]; however, achieving these effects requires precise control of optical feedback and facet reflectivity, making it a challenging task.
Beyond high modulation bandwidth, sufficient output optical power is equally paramount for reliable data transmission in fiber optic communication systems [
14]. Adequate output power directly enhances signal strength, extends transmission reach, and improves the signal-to-noise ratio (SNR), thereby reducing bit error rates (BERs) and ensuring data fidelity. However, achieving high output power is often constrained by material and structural factors, typically demanding bias currents exceeding 100 mA [
6,
12].
Traveling wave semiconductor optical amplifiers (TW-SOAs) exhibit quasi-high-pass filter characteristics in gain-saturated states [
15,
16], with a sharp decline in the low-frequency response and a relatively flat high-frequency response. It should be emphasized that the term “frequency response of the SOA” in this work does not refer to the intrinsic direct modulation capability of the SOA itself. Instead, it describes the dynamic gain response of the SOA—biased in the gain saturation regime—to intensity-modulated optical signals from the DFB section, with its gain varying differently for different modulation frequencies.
This behavior is primarily driven by gain saturation and carrier lifetime effects. Low-frequency modulation signals deplete carriers significantly, leading to reduced gain, while high-frequency signals quickly replenish carriers, maintaining higher gain. The gain G of the SOA in a saturated state can be expressed as follows [
17]:
where
represents the carrier lifetime,
f is the modulation signal frequency,
j is the imaginary unit,
denotes the saturated gain, and
and
represent the input optical power and saturation optical power of the SOA, respectively. When the optical power injected from the laser into the SOA is sufficiently high, the carrier population in the SOA becomes significantly depleted at low modulation frequencies, leading to pronounced gain suppression.
In contrast, at high frequencies, the modulation is too rapid for carriers to respond within each cycle, allowing the gain to be largely preserved. This asymmetric gain behavior—suppression at low frequencies and retention at high frequencies—results in a characteristic quasi-high-pass filtering effect. The magnitude of the frequency-dependent gain can be described by the following expression:
It is worth emphasizing that although the modulation signal is applied solely to the DFB section, and the SOA is not directly electrically modulated, the optical signal incident on the SOA carries intensity modulation with multiple frequency components. As a carrier-injection-based optical amplifier dominated by stimulated emission, the SOA exhibits a frequency-dependent dynamic response due to the interplay between carrier depletion and replenishment. Specifically, under low-frequency modulation, the slowly varying optical power leads to sustained carrier depletion over a longer duration, while the fixed electrical injection rate cannot replenish the carriers in time, resulting in a noticeable drop in gain. In contrast, at higher frequencies, the input optical power fluctuates rapidly within shorter time intervals, during which the carriers are not fully depleted. The injection current is sufficient to maintain a quasi-steady-state carrier concentration, thereby sustaining the optical gain. This frequency-dependent nonlinear response enables the SOA, when operated in the gain saturation regime, to function as a passive quasi-high-pass filter, selectively enhancing the high-frequency components of the modulated signal and effectively reshaping the overall system modulation response.
To address the challenge of achieving both increased bandwidth and high-power output while simplifying fabrication, we propose a DML design featuring a partially corrugated grating (PCG) integrated with a SOA. Simulation shows that the PCG design effectively mitigates random phase variations at the rear facet [
18], enhancing mode stability. Experiments indicate a side-mode suppression ratio (SMSR) exceeding 44 dB, validating the grating’s role in maintaining stability. Utilizing the SOA’s quasi-high-pass filter characteristics, the fabricated DML achieves a bandwidth of 25 GHz, representing a 7 GHz enhancement compared to the 18 GHz bandwidth of non-integrated SOA lasers. The laser outputs over 25 mW at a 60 mA bias current. In back-to-back (BTB) 26.5625 Gbps NRZ transmission tests under limited receiver bandwidth, clear eye diagrams were obtained, and TDEC comparisons show the fabricated DML has superior signal transmission capabilities.
Table 1 compares the proposed design with other monolithically integrated SOA-based DMLs, highlighting differences in modulation bandwidth and output power performance.
2. Device Structure and Fabrication
The structure of the proposed DML is shown in
Figure 1 and primarily consists of the distributed feedback (DFB) modulation region and the SOA region, each with a length of 150
m. The rear facet near the modulation region has a high-reflectivity coating (>95%), while the output facet near the SOA features a 0.5% reflective anti-reflective coating. A 10
m-wide, 200 nm-deep isolation region is etched between the modulation and SOA regions to ensure electrical isolation. The laser utilizes an identical active layer (IAL) structure [
18,
20,
21] with a partially corrugated grating design [
18]. A grating with a coupling coefficient
of approximately 7788 m
−1 is etched in the modulation region, positioned solely on the SOA side, leaving an unetched waveguide section between the high-reflectivity facet and the grating. This design effectively reduces the impact of random phase variations, enhancing mode stability.
To enhance the structural integrity and reproducibility of the device, the fabrication process and key parameters of the DFB-SOA integrated laser are further elaborated. The device is based on an InP substrate with an InGaAsP multiple quantum well (MQW) active region, designed for 1310 nm emission. The fabrication begins with epitaxial growth of the active and grating confinement layers, followed by grating etching and subsequent growth of the cladding and Ohmic contact layers. A uniform active layer is used across the entire wafer, with the DFB and SOA sections completed in a single epitaxial step without regrowth. This enables a monolithic integration with continuous structure and functional separation. A 2.4
m-wide ridge waveguide is formed in both sections via dry etching. In the DFB section, partial periodic gratings are defined using electron beam lithography and shallow etching (approximately 50 nm), yielding a grating coupling coefficient of
to balance feedback and modulation bandwidth. No grating is etched in the SOA section, which is used solely for optical amplification. Electrical isolation between the DFB and SOA sections is achieved by a 10
m-wide, 200 nm-deep trench fabricated via wet etching. This trench effectively blocks carrier diffusion and enables independent control of current injection. Ti/Pt/Au trilayer electrodes are deposited and patterned via a lift-off process to separately bias the DFB and SOA regions. A
passivation layer is deposited for surface protection, and metal pads are defined for subsequent electrical testing.
Figure 1b shows the top-view microscope image of the fabricated laser chip.
3. Device Characterization
Figure 2a shows the power–current (P–I) characteristics of the DML with and without an integrated SOA. The pink curve serves as the baseline for the DML without SOA, while the other curves illustrate the performance with SOA under varying SOA currents
. The slope, indicative of modulation efficiency, is determined by the modulation current
. Power saturation becomes evident as
approaches 70 mA, resulting in a reduced slope efficiency and limiting further optical power scaling. Without SOA, the optical power remains below 20 mW; however, with SOA, it increases to 28 mW at
mA and
mA, and scales further with increasing
.
Figure 2b depicts the relationship between SOA gain and both
and
. Notably, at
mA, the gain plateaus for
mA, indicating SOA gain saturation and limiting amplification capacity. In this saturation regime, the increase in carrier density, driven by higher
, becomes proportional to the increase in photons involved in stimulated emission. Consequently, for a constant
, the optical power increase in the linear region is directly proportional to the increase in
.
The SOA consumes approximately 36.84 mW under a driving current of 30 mA and a voltage of 1.228 V. In this device, the SOA section is relatively short (150
m) and exhibits strong optical confinement, resulting in a low saturation power (
), which makes it prone to entering the saturation regime. As shown in
Figure 3, under a fixed DC bias applied to the DFB section, the SOA gain initially increases rapidly with increasing injection current above threshold, but then saturates and even slightly decreases. This behavior reflects a typical static gain compression effect, indicating that the SOA can easily reach gain saturation even under relatively low input optical power. This static behavior lays the foundation for the frequency-dependent dynamic gain response under modulation. As further confirmed by the small-signal frequency response shown in
Figure 3, the gain compression is more pronounced at low modulation frequencies, while the gain remains higher at high frequencies. This results in a typical quasi-high-pass filtering characteristic exhibited by the SOA.
To understand the SOA’s impact on the DML’s dynamic performance, we measured the small-signal
response of both the SOA region alone and the integrated DML, as shown in
Figure 3 and
Figure 4, respectively. The measured
response of the SOA region (
Figure 3) exhibits characteristics consistent with a standalone SOA [
15], confirming the quasi-high-pass filtering effect introduced by the SOA integration.
For comparison, DMLs without the SOA region but with identical DFB parameters, were fabricated.
Figure 4e shows the frequency response of the DFB laser without SOA integration. A noticeable dip appears in the low-frequency region (2–12 GHz), and the overall curve is not flat. The
dB bandwidth is approximately 18.5 GHz, indicating a performance limitation caused by factors such as modulation efficiency and carrier recombination rate.
Figure 4a–d present the frequency responses of the DFB-SOA integrated laser under different DFB bias currents. In each subfigure, multiple curves correspond to various SOA bias currents (
), including the case where the SOA is inactive (
) and active (
). It can be observed that for all DFB bias conditions, the curves at
exhibit a similar low-frequency dip as seen in
Figure 4e. However, as
increases, the frequency response undergoes a significant improvement: the dip in the low-frequency region is gradually “filled,” and the mid- to high-frequency response is enhanced, resulting in an increased
dB bandwidth. For example, in
Figure 4d, under a DFB bias current
mA, the device achieves a
dB bandwidth of 25.8 GHz when
mA, which represents a 7.3 GHz enhancement compared to the case without SOA. This demonstrates a substantial improvement in modulation performance enabled by the integrated SOA.
The observed bandwidth enhancement originates from the quasi-high-pass filtering effect exhibited by the SOA under saturated conditions. When the modulated optical signal from the DFB section is injected into the SOA, the carrier depletion in the SOA leads to gain compression for low-frequency components, while the gain for high-frequency components remains relatively higher due to the delayed carrier response. As a result, the SOA provides a frequency-dependent dynamic gain response that reshapes the output signal of the DFB.This frequency-selective amplification mechanism effectively suppresses the pronounced low-frequency components in the DFB’s original response and simultaneously enhances the high-frequency modulation capability. Consequently, a “low-frequency suppression–high-frequency preservation” spectral shaping effect is formed at the output of the integrated device. Since the SOA exhibits stronger gain response at higher frequencies, it compensates for the amplitude roll-off in the high-frequency region of the DFB’s frequency response. This compensation shifts the overall dB bandwidth of the combined DFB-SOA system toward higher frequencies, thereby significantly improving the modulation bandwidth of the device.
The bandwidth variation trend is shown in
Figure 4f. For a fixed
, the bandwidth increases with rising
, attributable to the enhanced amplification from the SOA, which boosts output power and strengthens the quasi-high-pass filter effect, thereby improving frequency response and increasing bandwidth. When
is constant, the bandwidth initially increases and then decreases as
rises, peaking at
= 60 mA. As
approaches 65 mA, the laser power saturates, and further increases push the laser out of its linear operating region, resulting in reduced bandwidth.
At 25 °C and
30 mA, the spectrum of the DML under varying
conditions is shown in
Figure 5, featuring a side-mode suppression ratio exceeding 44 dB. Simulation results indicate that the PCG structure effectively mitigates the impact of random phase variations at the rear facet [
20], enhancing the mode stability of the laser. These findings are further validated by experimental results.
To assess the transmission performance enhancement enabled by the SOA, we conducted 26.5625 Gbps NRZ signal transmission experiments in back-to-back (BTB), 2 km, and 10 km configurations, evaluating the TDEC (Transmitter and Dispersion Eye Closure) and capturing corresponding eye diagrams, as presented in
Figure 6. TDEC is a standardized metric used to evaluate the signal quality at the transmitter side. It reflects the eye diagram opening after considering the effects of dispersion, filtering, and other transmission impairments. A lower TDEC value indicates better signal clarity and tolerance at the receiver, directly impacting the bit error rate (BER) performance. Therefore, TDEC serves as an important reference for assessing the overall system transmission quality.
Figure 6a demonstrates that, for a given filter bandwidth, the DML with SOA consistently exhibits a lower TDEC than the DML without SOA. This indicates improved signal quality and reduced signal degradation due to SOA integration. As shown in
Figure 6b, TDEC reaches its maximum at
and rapidly decreases as
increases to 10 mA for a fixed filter bandwidth. This improvement is attributed to the increased optical power from the SOA, which enhances the signal-to-noise ratio (SNR), leading to a clearer distinction between logic levels and a larger eye opening, as visually confirmed by the eye diagrams in
Figure 6. The clear eye diagrams after 2 km and 10 km transmission highlight the potential of the SOA-integrated DML for passive optical network (PON) applications. In contrast, the DML without SOA exhibits eye diagrams with broader transition regions, frequency chirp, and asymmetric edge slopes, indicative of significant chirp. For the DML with SOA, when the SOA is inactive (
), carrier transport minimizes the modulation impact on the DFB section, reducing chirp. Activating the SOA (with
) reduces carrier transport, increasing the DFB section’s modulation sensitivity and potentially leading to increased chirp, although the improved SNR and power still result in overall better performance as evidenced by the TDEC and eye diagrams.
It should be noted that, in theory, the dynamic gain response of the SOA may introduce a certain degree of chirp. However, in our experiments, no significant performance degradation attributable to chirp was observed. As shown in
Figure 6b, the TDEC value continuously decreases with increasing SOA bias current, and the eye diagram in
Figure 6c also exhibits a clear improvement in opening. These results indicate that the overall signal quality is effectively enhanced. Therefore, under the current device structure and bias conditions, the output power gain and signal amplification provided by the SOA play a dominant role in performance improvement. The positive effect of increasing the signal-to-noise ratio (SNR) clearly outweighs the potential negative impact of chirp.