1. Introduction

As the use of photonic integrated circuits (PICs) to deliver higher data transmission rates for optical networks increases, the thermal management in these highly integrated devices is becoming increasingly challenging. This is driving designers to consider more highly integrated thermal management approaches beyond the current standard macroscopic thermoelectric cooling (TEC) and conductive heat transfer within PIC packages [1,2]. In particular, the effect of thermal crosstalk in PICs, whereby a component’s temperature is influenced by a changing power dissipation profile in a neighbouring device(s), is of general interest in order to guarantee the desired performance of devices in the PIC, where the majority of photonic components are highly temperature sensitive. Semiconductor laser arrays in communications applications can be particularly challenging since there are conflicting requirements for low thermal resistance and adequate wavelength tuning through temperature control. Low global thermal resistance is desired to reduce the operating temperature of the laser array and remain within the performance envelope of commonly employed thermoelectric coolers, minimising the power consumption of the inefficient TEC. Wavelength tuning is typically provided by thermal means, however, where the device waveguide is heated using an integrated electric heater, but in order to minimise the power dissipated in this heater to achieve sufficient tuning, high thermal resistance is required in each laser.

Sato and Murakami investigated thermal crosstalk in InP-based DFB laser array chips mounted on a temperature controlled (Peltier) copper stage by measuring the shift of peak wavelength in an operating device when additional lasers in the array were turned on [3].They demonstrated that the spacing between devices determines the nature of thermal crosstalk between lasers. For lasers with a pitch spacing < 300 μm, lateral heat spreading dominated the thermal crosstalk behaviour. For a pitch spacing > 300 μm, the thermal crosstalk was dependent on the thermal performance of the heat sink, i.e., the global temperature increase across the chip. In this latter thermal crosstalk regime, they demonstrated the benefit of high thermal conductivity heat sinks; measuring a 1/3 decrease in crosstalk for a diamond substrate compared to silicon. Three general approaches can be used to reduce the array-to-coolant thermal resistance; increasing the thermal conductivity of the chip submount to improve lateral heat spreading, incorporating microchannel cooling into the chip submount and aggressively reducing the thickness of the chip substrate material [4]. In the context of cooling high-power laser-diode arrays, Goodson et al. [4] proposed a design incorporating all three approaches to minimise the device thermal resistance. The cooling system was comprised of a microchannel heat sink made of chemical-vapour-deposited (CVD) diamond, whose high thermal conductivity increases the efficiency of the channel-wall fins and reduces the array-to-coolant thermal resistance using a simple model for the combined conduction and convection problem. The resistance is calculated to be 75% less than that for a conventional configuration using a silicon microchannel heat sink.

While several approaches can be used to significantly reduce thermal crosstalk by reducing the overall device thermal resistance, in applications employing wavelength temperature tuning, this penalizes the wavelength tuning power performance. Klepser and Hillmer investigated the interplay between tuning and crosstalk for WDM applications where a strip heater is provided adjacent to each laser [5]. While they showed a reduction in crosstalk by reducing the thermal resistance of the chip/submount setup (by introducing a thermal grease interface layer), they also showed that just reducing the thermal resistance of the device/submount does not always result in reduced crosstalk. Their combined experimental/simulation analysis concluded that placing a thermal barrier close to the active region in the device (between it and the substrate), reduces the thermal crosstalk because the tuning efficiency was now greatly increased. Therefore less power is applied to each device’s strip heater resulting in less thermal power dissipation in the chip.

Gilardi et al. investigated the thermal performance of InP-based PICs and demonstrated the effect of substrate thickness and deeply etched trenches (thermal barriers) on reducing the thermal interplay between active and passive components [6,7]. Through chemical wet etching, deep trenches were introduced between semiconductor optical amplifiers (SOA) and Mach-Zender (MZ) modulators on a PIC, with 100 μm deep trenches placed midway between the SOA (applied current 100 mA) and MZ, and a significant reduction in the change of optical extinction ratio as compared to co-located devices without a trench.

In this work, we study the lasing behaviour of a thermally tuned slotted laser array with integrated SOA. The laser array is intended to operate as a flexible wavelength source in dense wavelength division multiplexing (DWDM) applications such as metro systems with channel spacing of 200 GHz. The primary design goal is to optimise the cooling strategy by removing or better utilising the inefficient macroscopic TEC, while maintaining tight control of the transmitting lasers wavelength within an allowable value of ± 25 GHz for these systems. For dense laser arrays an efficient cooling strategy is being developed to minimise a transmitting lasers wavelength shift due to thermal cross talk from neighbouring lasers being turned on in preparation for a channel change, while maximising the tuning efficiency of the transmitting laser. We first experimentally characterize the optical and thermal performance of the laser array. This information is then used to calibrate a numerical model used to explore the performance implications of several thermal design changes based on modifying the thermal characteristics of the laser chip submount. We find that decreasing the submount thermal resistance reduces thermal crosstalk, but not below the desired level. A submount design with embedded microfluidics is proposed that reduces thermal crosstalk to less than 2 K (< ~25 GHz). This microfluidic approach will allow for better utilisation of the macro TEC in our package as the heated fluid can now be transported to any region on the board for secondary heat exchange, providing significant possibilities for reduced size and increased integration in PICs.

2. Laser array design

The schematic structure of the slotted single mode laser array is shown in Fig. 1. Each laser of the array has a typical 2.0 μm-wide surface ridge waveguide structure. The laser epitaxial structure is based on a standard 1550 nm LD design. The active region consists of five AlGaInAs quantum wells. Above it are 1.6 μm-thick p-doped InP layer, 50 nm-thick p-doped InGaAsP layer, and 200 nm-thick InGaAs contact layer. One side of each laser has multiple uniformly distributed slots that act as an active DBR reflector of the laser and provide sufficient one side (front) feedback for lasing operation. The other (back) side reflection is provided by the facet. Because in this case we still need to cleave the back facet, there is a single mode issue caused by the uncertainty of this cleaving position. To remove this yield problem, each laser is divided into two sections electrically isolated by the last slot from the group of slots. As seen in Fig. 1, the front section includes a group of slots and the back section consists of a straight waveguide section. By tuning the back section current of some lasers in the array, i.e. tuning the longitudinal mode position relative to the reflection peak, it is possible to make a laser array with good single mode performance for all the channels. Considering the loss caused by the etched slots, a SOA is naturally integrated with but electrically isolated from each laser to boost the output power. To reduce the reflection from the front facet, the SOA section is curved to generate a 7° angle from the normal of the facet. To further improve the laser performance in terms of increasing the output power and reducing the threshold current, antireflection (AR) and high reflection (HR) coating films are applied to the front and the back facets of the laser array, respectively.

 

Fig. 1 Schematic outline of the 10-channel slotted laser array with inset a SEM image of the fabricated slots and the cross-section outline of the device.

Download Full Size | PDF

For each laser in the array, the slot width, depth, and number are key geometric design parameters that need to be optimized. A 2D scattering matrix method was used for this optimization [8,9]. To make the laser fabrication compatible with standard photolithography, the minimum slot width of approximately 1.1 μm was specified. The slot depth was optimized to be 1.35 μm, approximately 73% of the ridge height. For such a slot depth a single slot can provide around 1% and 97% amplitude reflection and transmission, respectively. Because the reflection from a single slot is quite weak, a group of slots are used to provide sufficient feedback for the lasing operation. A characteristic slot period of approximately 9 μm was used to obtain a high reflection from the slots such that they act as high order surface gratings with 37th grating order. An optimal slot number of 24 was selected based on the trade-off between maximizing the reflectivity while minimizing the bandwidth of the reflection peaks and keeping the cavity length short, resulting in a calculated amplitude reflection and transmission of around 0.43 and 0.4, respectively. Such a group of slots ensure that the reflection spectrum has a narrow bandwidth to achieve a good single mode operation with minimum threshold. The laser array was designed to have a frequency spacing of 400 GHz corresponding to 3.2 nm wavelength spacing, which is achieved by controlling the slot period according to

 

Fig. 2 Individual laser channel characterisation. (a & b) Continuous wave (CW) LIV curves for laser channels. 1 – 10 on the laser bar performed at a copper heat sink temperature of T_hs = 20 °C, (c) measured spectral output from the ten channels showing the C-band coverage at a heat sink temperature of 20 °C (d) Wavelength temperature coefficient dλ/dT for laser device 5 in the array.

Download Full Size | PDF

3. Experimental

3.1. Fabrication and individual laser characterization

The designed laser array was fabricated following our previously described process [9–11] with a ridge spacing of 360 µm. Briefly, two steps of inductively coupled plasma (ICP) etching with Cl₂/N₂ gas combinations were used to form the ridge and the slots. The total length of each laser in the fabricated array is around 400 μm integrated with the 190 μm long curved front SOA. The back section of the laser was about 185 μm long between the back facet and the first slot (from left). The front section was about 205 μm long slot section. After the ridge was passivated and metal contacted (the electrode metal covered the slots area except for the isolation slots), the lasers were coated with high reflection (HR) and antireflection (AR) films. Finally, the laser arrays were cleaved into individual bars and eutectically bonded to AlN carriers. The thicknesses of InP substrate and of the carrier are 120μm and 620µm, respectively.

The mounted device was placed on a copper heat sink temperature controlled using a thermoelectric cooler (TEC). A thermistor embedded in the copper heat sink below the mounted device was used for feedback control. To ensure good thermal contact between the mounted device and the heat sink, a 50 μm thick layer of thermal grease (Aremco Heat-Away 641-EV) with a stated thermal conductivity of k_g = 5.58 W/m.K was applied to the copper heat sink using the doctor blade technique. First, we electrically connected the front section and back section of each laser together and measured the light-current-voltage curves of the lasers under continuous-wave (CW) conditions at a copper heat sink temperature of 20 °C. To remove any FP cavity influence, the SOA was left unbiased. Figure 2 shows the measured LIV curves and spectral output for all of the 10 channels within the array. The HR coated back facet produce a high reflection which results in a low threshold current of about 18~20 mA. The LI curve of device 6 demonstrates a kink which can be attributed to mode hopping introduced by the cleaving position of the back facets [10,11].

A spectroscopic technique was implemented to characterize the individual laser thermal resistance, $R_{th}=\left(T_{ar}-T_{hs}\right)/P_{diss}$, where T_ar, T_hs and P_diss are the active region temperature, heat sink temperature and dissipated power, respectively. Output light from the lasers was collected by a lensed single mode fiber and analyzed by an Agilent 86140B optical spectrum analyzer (OSA). The resolution of the OSA was set to be 0.06 nm with sensitivity of −80 dBm. Two sets of measurements were performed. The first set of measurements was used to establish a baseline for the shift in lasing wavelength as a function of active region (heat sink) temperature (dλ/dT). This experiment was performed under pulsed wave (PW) conditions (pulse duration = 1 µs, duty cycle = 0.1%) to minimize device self-heating in the active region. Results for the pulsed mode wavelength shift as a function of active region temperature (dλ/dT) are shown in Fig. 2(d) for laser device 5. Over the entire array of lasers we found a mean of (dλ/dT)_m = 0.0938 ± 0.006 nm/K. Thus, in order to cover the free spectral range (FSR) of each tunable slotted laser, FSR ~3.2 nm [11], a temperature change of ~34 K is required. The second set of measurement was performed under CW conditions to measure the shift in wavelength as a function of the applied electrical power to the laser (dλ/dP). The thermal resistance was then determined as

3.2. Laser array thermal crosstalk

Next, we experimentally studied the effect of thermal crosstalk in a slotted laser array. The optical output from laser device 5 in the centre of the array was studied; specifically its peak wavelength shift to determine the relative temperature change in the active region of the device. To measure the thermal crosstalk between devices in an array we measured the wavelength shift in laser 5 when another device on the chip was operational subject to an applied current of 100 mA (CW). The output signal from the device 5 was collected through a lensed SMF and the spectrum was recorded using an optical spectrum analyser. The change in active region temperature (ΔT) of device 5 as a function of which neighbouring laser was operational is shown in Fig. 3. When device 1 was turned on, the peak wavelength of device 5 is shifted by 0.15 nm corresponding to ΔT = 1.56 K. The neighbouring lasers 4 & 6 caused larger temperature changes of 2.7 K and 2.86 K, respectively, when operational. The same experiment was performed without the thermal grease layer between the copper heat sink and the AlN submount to investigate the effect of increased interface resistance between the chip and heatsink. Now when device 1 was turned on, the peak wavelength of device 5 is shifted by 0.32 nm corresponding to a temperature rise of ΔT = 3.3 K, a larger value than for lasers 4 & 6 with the thermal grease interface layer. The neighbouring lasers 4 & 6 caused correspondingly larger temperature changes of 4.79 K and 4.81 K, respectively, when operational. This large increase in thermal crosstalk highlights the important role of the overall heat sink thermal resistance in our system. To place the observed wavelength change in a communications context, it is relevant to point out that every 2 K temperature change corresponds to a frequency change of 25 GHz in the C-band. Thus, the observed wavelength shifts are significant in terms of DWDM applications where channel spacing is typically ≤ 100 GHz [12].

 

Fig. 3 Change in peak output wavelength of laser 5 as a function of neighbouring operational laser under 100 mA (CW) operation. Data was measured for the chip placed directly on the heatsink or with a grease thermal interface layer.

Download Full Size | PDF

4. Numerical model

To explore the impact of thermal design changes on crosstalk in the laser array, we developed a multiphysics numerical model of the laser array. The numerical model of the laser array considers the coupling temperature and electric fields over the entire ridge laser structure shown in Fig. 1. To numerically capture all the characteristics of the laser, a phenomenological model of laser operation was considered. In particular, above the threshold current, I₀, the voltage across the active region is clamped at the ideal diode voltage, V_d, plus a current independent series voltage, V_s [13]. The dissipated electrical power, P_D, over the entire laser then consists of the ohmic behaviour outside the active region and the non-ohmic behaviour within the active region minus the optical output power, P_o, expressed as in [13] where, I, is the applied current and, R_s, the device series resistance.

 

Fig. 4 Schematic cross-section of the simulated laser structure (not to scale). The boundary condition is highlighted (dashed line).

Download Full Size | PDF

Table 1. Thermal and electrical properties of the materials present in the laser structure found in [14,15].

View Table

To determine the series resistance and clamping voltage values for use in our model, the CW voltage-current (V-I) characteristics of the slotted lasers were analysed. For laser device 5, the clamping voltage, evaluated from the intercept of the V-I curve above the lasing threshold, was found to be V_d + V_s = 0.8865 V. A series resistance of R_s = 9.2 Ω was found by evaluating the slope of the same V-I curve. To match the experimentally determined series resistance, the contact resistance in the p region was specified as 1.043 × 10⁻⁹ Ω-m². In order to capture the heat generation associated with the active region ($I\left(V_d+V_s\right)-P_o$), P_o was evaluated directly from the experimental data [Fig. 2(b)] assuming 100% reflection from the back facet. To verify the predictions of the numerical model, the electrical/optical characteristics of a slotted laser was simulated over an applied current range from 20 to 150 mA and compared to measured values. Figure 5 shows the electrical potential and thermal resistance of laser device 5 along with the characteristics simulated by our model. The simulated electrical potential matches closely the measured values. The thermal resistance was measured at one applied current (100 mA) with a value of 72.55 K/W found. The thermal resistance was simulated to vary from 74.3 - 70.9 K/W over the 25 – 150 mA applied current range, in close agreement with the measured value.

 

Fig. 5 Measured and simulated current-voltage characteristics of laser device 5 at 20°C with the simulated and measured thermal resistance of the device also shown.

Download Full Size | PDF

4.1. Comparison of numerical predictions and experimental measurements of thermal crosstalk

Next we compared our numerical predictions against the experimental crosstalk behaviour. The temperature rise in laser 5′s active region as a function of the operation of neighbouring lasers was numerically simulated and compared with measurements, as shown in Fig. 6(a) and 6(b). The current applied to the neighbouring laser was over a 30 – 150 mA range with measurements also taken within this range. Figures 6(a) and 6(b) show that the temperature rise in laser 5 is proportional to the applied current at the neighbouring laser. The increase was approximately linear revealing the ratio of dissipated power to optical power does not greatly change in the neighbouring laser with applied current.

 

Fig. 6 Comparison of the simulated (lines) and measured (symbols) temperature rise in the active region of device 5 as a function of operating laser and applied current (a) lasers 1-4 (b) lasers 6-10.

Download Full Size | PDF

Figure 7 compares the measured and simulated thermal resistance owing to thermal crosstalk. Here R_th,cross = ΔT₅/P_D,n, where, ΔT₅, is the temperature rise in laser 5 owing to the heat dissipated in a neighbouring laser n, P_D,n. We found good agreement between our simulations and experimental data. It is shown in Fig. 6 that thermal crosstalk is inversely proportional to the laser distance and eventually levels off. This levelling off, rather than a continual reduction towards zero, suggests a spacing between individual lasers exists where, rather than lateral heat spreading, the thermal cross talk is primarily due to vertical heat transport and a subsequent rise in the global temperature of the substrate in agreement with the findings of Sato & Marakami [3]. This is further revealed by the difference between the temperature rise for laser 5 caused by the operation of the furthest lasers [Figs. 6(a) and 6(b): 1, 2, 9 and 10] is negligible across the entire applied current range.

 

Fig. 7 Comparison of the simulated (NUM) and experimentally measured (EXP) crosstalk thermal resistance (with grease applied between the chip and heat sink) as a function of operational neighbouring laser with an applied current of 100 mA (CW).

Download Full Size | PDF

In summary, the process of fitting the measured data to our model involved first measuring the I-V and L-V curves of the lasers in the array, separately building a multi-physics model in COMSOL using all known material parameters, varying the MQW conductivity (typically an unknown) to fit the electrical data to the model and determining h to fit the thermal data. The model will need to be optimised and calibrated for different laser array chips with some knowledge of the expected range of unknown values, but we have shown this is a relatively straightforward and accurate method.

4.2. Simulating alternative substrate materials

The analysis of crosstalk within the current laser array revealed the role played by the substrate where the heat generated in devices furthest from laser 5 travelled vertically to the substrate, with a subsequent general rise in substrate temperature that feeds back into all devices on the chip. To investigate the effect of the chip carrier thermal characteristics on thermal cross talk, the material properties of diamond (k_DIA = 2000 W/mK [16]) and anisotropic highly-ordered pyrolytic graphite (HOPG) (kHOPG_x: 6.12 W/mK, HOPG_y: 1290 W/mK [17]) were simulated as the chip carrier material to investigate how the thermal conductivity of the substrate influences this transport phenomenon. The carrier thickness of 620 μm was maintained to allow comparison with the measured data and HOPG was considered orientated in both a vertical (vHOPG) and horizontal (hHOPG) orientation. Figure 8(a) shows the temperature rise varies as a function of laser number at an applied current of 100 mA (CW). It shows that all the curves are slightly asymmetric at laser 5. For the isotropic materials, in general, the carrier layer with high thermal conductivity (Diamond) reduces thermal crosstalk below the values achieved by AlN due to decreasing the thermal resistance of the entire setup. The reduction in thermal crosstalk with increased laser spacing is less pronounced than for AlN, with the operation of lasers 9 & 10 showing higher thermal crosstalk. This is due to the higher thermal conductivity material facilitating a more even temperature rise across the whole chip as the heat more readily travels laterally. In these cases the heating of device 5 is primarily related to the heating factor in the heat sink.

 

Fig. 8 Effect of submount material properties. (a) Comparison of the simulated temperature rise in the active region of laser 5, as a function of neighbouring laser operated at 100 mA (CW) applied current, with different materials considered for the submount. (b) Laser device 5 thermal resistance for different submount materials.

Download Full Size | PDF

For the anisotropic HOPG, the thermal crosstalk for the furthest lasers is reduced lower than for isotropic materials. However, for the nearest laser there is a larger increase in temperature in laser device 5 due to reduced lateral heat spreading associated with the significantly reduced cross plane thermal conductivity (k ~6 W/m.K). Changing the orientation of the HOPG basal plane such that it is perpendicular to the primary heat flow path has a drastic effect on crosstalk, resulting in temperature rise in device 5 of larger than 10 K during operation of any other lasers in the array. Figure 8(b) shows the corresponding change in the thermal resistance of laser device 5 with the choice of submount. The thermal resistances for a single laser are AlN - 74.61 °C/W, DIA - 67.83 °C/W, vHOPG - 83.24 °C/W, hHOPG - 151.26 °C/W.

4.3. Microfluidic enhanced substrates

Although diamond reduces the crosstalk between individual devices to ~2 K, in order to consistently achieve the desired wavelength control in tunable laser structures we deem the use of solid-state submounts as inadequate. We modelled the effect of embedding microfluidic channels in the substrate to increase the heat transfer away from the active region of the array as shown in Fig. 9(a). The model considered a 200 µm x 200 µm channel that extends along the complete length of each lasers ridge, embedded with the top channel wall 30 µm below the top surface of the AlN carrier [Fig. 9(b)]. Therefore the primary thermal resistance between the laser and channel considered was the InP substrate. Using the method in [18] we estimated a heat transfer coefficient of 5x10⁴ W/m²K and applied this as the boundary condition at the microchannel walls.

 

Fig. 9 (a) Simulated temperature rise in the active region of laser 5, as a function of neighbouring laser operated at 100 mA (CW) applied current corresponding to the AlN submount used in our experiments or a submount made of diamond or one with micro-fluidic channel cooling embedded in the substrate. (b) Simulated temperature profile of two operating lasers in the array when a microchannel fluid cooling solution is embedded in an AlN submount.

Download Full Size | PDF

The temperature increase in laser 5 remained below the acceptable value of 2 K during the operation of any other laser on the chip. This would allow the operation of any two lasers on the chip during switching. While this configuration reduces the thermal crosstalk between devices below the acceptable level of 2K, the thermal tuning of individual devices must also be considered. We simulate the thermal resistance of a laser in an array with microfluidic channels embedded in the substrate will be 74.4 K/W. This is marginally below the value simulated for an AlN submount and greater than a diamond one. Therefore the use of a microfluidic enabled submount will not result in a significant loss in tuning performance of the laser array.

5. Conclusions

We have experimentally and numerically studied the thermal crosstalk and tuning behaviour in a tunable wavelength slotted laser array designed for use as a flexible C-band wavelength source in DWDM applications. Our experiments have shown that the standard thermal solution with the laser array chip mounted on an AlN submount does not allow for wavelength stability of better than ~0.2 nm corresponding to an allowable ± 25 GHz in the C-band when neighbouring devices (spacing = 360 µm) are operated at an applied current of 100 mA (CW). Using our experimentally validated numerical simulations, we explored the effect of changing the submount thermal characteristics. We found that, with its larger thermal conductivity, diamond has an increased effect on suppressing cross talk and marginally decreases the wavelength tuning power coefficient. The use of vertically aligned anisotropic HOPG reduces crosstalk for lasers furthest from the operating device but increases it for neighbouring devices due to the low thermal conductivity in the horizontal direction. An alternative submount with embedded microfluidics was shown to suppress cross talk below an allowable value of 2 K for all devices on the chip. The critical role submounts play in the thermal performance of an array of tunable lasers has been explored in detail. We have shown that high thermal conductivity submounts on their own are not enough to control wavelength stability in optical networks. The use of microfluidic substrates will allow us to increase photonic integration, reduce the size of and more optimally use inefficient thermoelectric coolers saving on package real-estate and cost.