1. Introduction

The rapid growth of the data traffic has been driving the need for larger capacity of optical communication systems and faster port speed of network elements. To accommodate rapidly increasing Internet data traffic, 40G/100G Ethernet was standardized in June 2010, and IEEE 802.3 is now working on next generation 100G Ethernet standardization. 4 × 25 Gb/s LAN-WDM (local area network - wavelength division multiplexing) method (100GBASE-LR4) was adopted as a 100G Ethernet standard for 10 km over single mode fiber. The center wavelengths of the four channels are 1295.56 nm (lane 0), 1300.05 nm (lane 1), 1304.58 nm (lane 2), and 1309.14 nm (lane 3) [1,2].

Due to the limited space in data centers, transceiver modules with small form factor are required to increase port density of line cards in high capacity routers and switches. One of the key issues for miniaturization is how to integrate various discrete components into one module, such as ROSA (receiver optical sub-assembly) module integrated with an optical DMUX (demultiplexer). In order to save packaging cost and time in assembling process of ROSA modules, it is important to realize high alignment tolerance between optical input signals and active areas of PDs (photodiode). Up to now, reported alignment tolerance has been in the range of only a few micrometer for the most 4 × 25 Gb/s ROSA with PLC (planar lightwave circuit) based optical DMUX [3,4]. We have developed a 4 × 25 Gb/s ROSA module for 100G Ethernet optical transceiver, and this ROSA module shows very large alignment tolerance of more than ± 250 µm between the optical DMUX and four PDs [5]. This is the largest alignment tolerance value among 4 × 25 Gb/s ROSAs to our knowledge.

To enhance the alignment tolerance, we proposed an optical DMUX composed of a parallelogram-shaped optic block and four thin film filters. We also used two-lens system with a collimating lens and a focusing lens. Since the optical path depends on the lengths and angles of the optic block, it is required to analyze dimensional tolerances as well as alignment tolerances of the optical DMUX. Furthermore, it is needed to define the fabrication specifications for the dimension of the optic block through the dimensional tolerance analysis.

Several thin-film filter based ROSA modules have been reported, and the only alignment tolerances were investigated as a function of coupling efficiency in order to estimate the tolerances for the passive or active assembly process [6–8]. In this paper, we report analysis results of the dimensional tolerance for the parallelogram-shaped optical DMUX used in our successfully developed 4 × 25 Gb/s ROSA module by using three-dimensional simulation. These analysis results can be also used for a reference in other thin-film filter based optical DMUX with zigzag multi-reflection structure.

2. ROSA structure and performance

The 4 × 25 Gb/s ROSA module is composed of three sub-blocks as shown in Fig. 1(a). An input collimator with LC receptacle is used for collimating the input optical signal launched from single mode fiber. An optical DMUX block separates four optical signals in different wavelengths spatially. A PD block converts the four demultiplexed optical signals to electrical signals. Figure 1(b) shows the functional block diagram of the ROSA.

 

Fig. 1 (a) Schematic diagram of the ROSA. (b) Block diagram of the ROSA. (c) Schematic diagram of the optical DMUX block.

Download Full Size | PDF

Figure 1(c) shows the schematic diagram of the optical DMUX block, and it consists of a parallelogram-shaped optic block and four LAN-WDM thin-film filters. The input and output regions of the optical signals are anti-reflection coated, and the opposite side of the filters in the optic block is formed by high-reflection coating. The tilting angle is 13.5° and the pitch of the demultiplexed optical signals is 2 mm.

The PD block is composed of four PIN-PD chips, four TIA (trans-impedance amplifier) chips, an AlN submount, a four-channel array lens, a rectangular-shaped TO (transistor outline) stem, and a flexible PCB (printed circuit board). To reduce the distortion and jitter of the output electrical signal, we optimized the data interface part of the TO stem and we have suggested various pad structures of flexible PCB [9,10].

The receiver sensitivities of the four lanes were measured below – 12.5 dBm of average received power with PRBS (pseudo-random bit sequence) 2³¹–1 data pattern, the extinction ratio of 11 dB, and BER of 10⁻¹² as shown in Fig. 2(a). The penalties were less than 0.1 dB in case that all four lanes are simultaneously input to the ROSA. Figure 2(b) shows the electrical output eye diagrams of the ROSA. A LiNbO₃ Mach-Zehnder intensity modulator was used to generate 25.78 Gb/s optical signal. The eye diagrams are clearly opened at input optical power level of – 10 dBm. The relative spectral response of the fabricated ROSA is plotted in Fig. 2(c). We obtained a 1 dB bandwidth of larger than 3 nm, and a high adjacent channel isolation of more than 30 dB for all lanes. The gray regions represent the wavelength range defined in IEEE 802.3ba standards. In addition, we measured the alignment tolerances of the four lanes under condition that all lanes were aligned optimally. Lateral alignment tolerances were measured more than ± 250 µm for all lanes, and tolerance for tilt angle of the beam incident on the PD block was ~ ± 0.25° [5].

 

Fig. 2 (a) BER performance of the ROSA at data rate of 25.78 Gb/s, data pattern of PRBS 2³¹–1. (b) Electrical output eye diagram of the ROSA at input power level of – 10 dBm. (c) Spectral response of the ROSA.

Download Full Size | PDF

3. Analysis of dimensional tolerance for the optical DMUX block

Since the deviation in the position of the demultiplexed beams is determined by the accuracy of the angle between two mirror planes and the accuracy of the beam path length in the optic block, it is required to define the allowable values for the angle and the dimension of the optic block. It is also needed to make it as accurate as possible according to the design specification without error. In order to reduce manufacturing error we used the precision glass molding process with the glass material, which was K-PG325 with refractive index of 1.495 at 1310 nm. The precision glass molding process has been reported as the method with high precision and low cost [11].

To analyze the dimensional tolerances of the optical DMUX block, we developed a program using MATLAB software. The program was designed by using Snell’s law and vector calculations in the three-dimensional Cartesian coordinate system [12].

In the interface between two material with different refractive indices ${\eta }_1$ and ${\eta }_2$, the direction vector of the transmitted beam t and the direction vector of the reflected beam r will be calculated by Eq. (1) and (2) from the direction vector of the incident beam i and normal vector n, orthogonal to the interface, and the angle of incidence ${\theta }_i$. In our simulation, ${\eta }_1$ is refractive index of air and ${\eta }_2$ is refractive index of the optic block.

Figure 3(a) is the diagram of the optic block used in the simulation, and shows the definitions of length (L) and angles (θ₁, θ₂, θ₃, θ₄). We have two assumptions for the simulation: (1) two mirror planes of the optic block are flat, and (2) input beam is parallel to the Y-axis as shown in Fig. 3(a). Design values of the optic block are listed in Table 1. Since PD and TIA of each channel were placed with 2 mm separation on the AlN submount of the PD block, the pitch of demultiplexed beams in the optic block should be 2 mm. For this reason, the values of length and angles of the optical block in Table 1 were deduced. Figure 3(b) illustrates the beam traveling paths through the optic block. Displacement is defined as the distance in XZ plane between the ideal beam path and the tilted beam path generated by error of length and angles at the beam exiting point of the optic block for Lane 0. In this analysis, displacement was simulated as a function of two parameters, such as (L versus θ₁), (L versus θ₃), (θ₁ versus θ₂), (θ₃ versus θ₄) and (θ₁ versus θ₃).

 

Fig. 3 (a) A diagram of the optical block for three-dimensional simulation. (b) Illustration of the beam traveling paths through the optic block.

Download Full Size | PDF

Table 1. Design Values Used in the Simulation

View Table

Figure 4(a) shows the simulation results of displacement as a function of the length L and angle θ₁ in the optic block. Since the minimum lateral alignment tolerance between the optical DMUX block and PD block was ± 250 µm, we set ± 250 µm as an allowable displacement value. Design values of L and θ₁ were 6.7 mm and 76.5° respectively. Tolerance of L was simulated to ~ ± 300 µm for the case of θ₁ = 76.5°, θ₂ = 103.5°, θ₃ = 90°, and θ₄ = 90°. It means the optic block is highly tolerant for the length. Figure 4(a) also shows there are many combinations of L and θ₁ to satisfy the condition of less than ± 250 µm displacement. Even though θ₁ is increased to larger than 76.5°, if L is decreased to less than 6.7 mm, displacement of the output beam will be less than ± 250 µm.

 

Fig. 4 Displacement as a function of (a) L and θ₁, (b) L and θ₃ in the optic block. Tolerance of L was simulated to ~ ± 300 µm for the case of θ₁ = 76.5°, θ₂ = 103.5°, θ₃ and θ₄ = 90°.

Download Full Size | PDF

Figure 4(b) is the simulation results of displacement as a function of L and θ₃ in the optic block. θ₃ is the angle between the input beam and incident plane of the optic block as shown in Fig. 3(a). In case of θ₃ = 90° and the other angles correspond to the design values, L has the large tolerance of ~ ± 300 µm. However, if θ₃ becomes larger than 90.1° or smaller than 89.9°, the displacement of the output beam will be higher than ± 250 µm regardless of L value. We can infer that θ₃ should be well controlled during the manufacturing process.

Figure 5(a) shows the displacement generated by variance of θ₁ and θ₂ in the optic block. The design values of θ₁ and θ₂ were 76.5° and 103.5° respectively. The angular tolerances of θ₁ and θ₂ were both simulated to ~ ± 0.1°. If the sum of θ₁ and θ₂ becomes ~180°, displacement is suppressed to less than ± 250 µm. The displacement is generated by errors of θ₁ and θ_2, and pitch of the demultiplexed beams is changed by tilting the optic block.

 

Fig. 5 Displacement as a function of (a) θ₁ and θ₂, (b) θ₃ and θ₄ in the optic block. The angular tolerances of θ₁, θ₂, θ₃ and θ₄ were simulated to ~ ± 0.1°.

Download Full Size | PDF

The displacement caused by variance of θ₃ and θ₄ in the optic block is plotted in Fig. 5(b). The design values of θ₃ and θ₄ are both 90°. The angular tolerances of θ₃ and θ₄ were both simulated to ~ ± 0.1°. When they are parallel to each other, the displacement stays within ± 250 µm.

The influence of θ₁ and θ₃ on displacement is shown in Fig. 6. According to the results, θ₁ and θ₃ are critical parameters that affect the displacement, and these values should be controlled within ± 0.1° during the manufacturing process.

 

Fig. 6 Displacement as a function of θ₁ and θ₃ in the optic block. The angular tolerances of θ₁ and θ₃ were simulated to ~ ± 0.1°.

Download Full Size | PDF

In order to validate the simulation, we calculated the displacement as a function θ₁ by rule of thumb. Figure 7 shows calculation procedure of the displacement in case there is an error in θ₁ in the optic block. Assuming that there is no error in the other parameters (θ₂, θ₃, θ₄ and L), the incoming beam travels only on the XY plane. The black line represents the ideal beam path, and the red line represents the beam path tilted by error of θ₁ ($\Delta {\theta }_1$). In Fig. 7, Lt calculated by Snell’s law is 6.5 mm, D₁ (displacement at the first point) becomes L_{t$\sin \Delta {\theta }_1$}. Since D₂ (displacement at the second point) is accumulated by D₁, D₂ becomes two times larger than D₁. D₃ (displacement at the third point) becomes 5 times larger than D₁, because displacement generated by error of reflection plane and displacement generated by error of the incident beam are added to D₂. In this way, D₇ (displacement at the final exit point) is to become 25 times larger than D₁, and it should be less than lateral tolerance of ± 250 µm.

 

Fig. 7 Calculation procedure of the displacements in case of being an error in θ1 in the optic block.

Download Full Size | PDF

By Eq. (3), we can get the result that error of θ₁ ($\Delta {\theta }_1$) should be equal to or less than ~0.09°. Since sin$\Delta \theta$ ≈$\Delta \theta$ for $\Delta \theta$ ≈0 and |$\Delta \theta$| = |$\Delta {\theta }_1$ + $\Delta {\theta }_2$|, Eq. (3) is able to be expressed the following generalized form.

Equation (4) is well matched to the results shown in Fig. 5(a), which are the exact solutions simulated by vector equations. The displacements generated by the errors of θ₃ and θ₄ ($\Delta {\theta }_3$ and$\Delta {\theta }_4$) can be also derived to the same equation as Eq. (4).

Since there is a possibility of tilting the input collimator in the practical alignment process, and the tilting effect changes the direction of the incident beam into the optic block, it is also required to investigate the influence of the tilt angle of the collimator on displacement of the optic block. Figure 8(a) shows the definitions of tilt angles ($\alpha \text{and}\beta$) of the input collimator. $\alpha \text{and}\beta$ represent the tilt angle of the collimator on the XZ and YZ planes, respectively. The displacement generated by variance of $\alpha \text{and}\beta$ in the collimator is shown in Fig. 8(b). The angular tolerances of $\alpha \text{and}\beta$ were simulated to ~ ± 0.6° under the condition of the lateral alignment tolerance of ± 250 µm for the PD block. However, since the allowable tilt angle of the incident beam on PD block is ~ ± 0.25°, the practical tolerance on tile angle of the collimator is limited to less than ~ ± 0.25° [5].

 

Fig. 8 Displacement as a function of the collimator tilt angle. (a) Illustration of tilt angles ($\alpha \text{and}\beta$) of the collimator. (b) Displacement against $\alpha \text{and}\beta$. The angular tolerances of $\alpha \text{and}\beta$ were simulated to ~ ± 0.6°.

Download Full Size | PDF

4. Experimental results

Experimental setup to measure the displacement and the dimensional errors of the fabricated optical DMUX block is shown in Fig. 9(a). Before starting the measurement, in order to make sure that the input collimator, the optical DMUX block and the PD block are placed in their own positions, we aligned these blocks to have the maximum monitoring current values for all lanes. We supplied power to the only one PD (lane 3) and measured the output current by moving PD block in the direction of X-axis.

 

Fig. 9 (a) Experimental setup for measuring the displacement and the dimensional errors of the fabricated optical DMUX block. (b) The measured results of the coupling efficiency along the X-axis. The peak values of the coupling efficiency represent output positions of the demultiplexed beams in the optical DMUX block.

Download Full Size | PDF

Figure 9(b) shows the experimental results of the coupling efficiency along the X-axis. Measured pitches of the output beams were 1960 µm, 2000 µm, and 2040 µm, respectively. Pitches were increased as the beam path length became longer. This increasing value of pitch means that there were errors of θ₁ and θ₂ in the mirror planes of the optic block and the sum of θ₁ and θ₂ was slightly larger than 180°. As shown in Fig. 9(b), lateral alignment tolerances were more than ± 250 µm for all lanes at coupling efficiency of 90% (0.5 dB loss). Larger alignment tolerances of the lane 1 and 2 were caused by the reason that alignment process was carried out based on lane 1 and 2.

In our ROSA structure, incident angle can be controlled by tilting the optical DMUX block, which makes the same effect as tuning θ₁ and θ₂. Although the pitch of the demultiplexed beams in the optical DMUX block is designed with 2 mm, since there are some errors of θ₁ and θ₂ in the actually manufactured optical DMUX block, the displacement is generated and pitches become larger than 2 mm in case that the sum of θ₁ and θ₂ is larger than 180°. The pitches increased by errors of θ₁ and θ₂ can be decreased by tilting the optical DMUX block toward the direction that the incident angle becomes smaller. One of the advantages of the ROSA is that the dimensional errors of the optical DMUX block can be compensated through the alignment process.

The displacement was calculated to ~230 µm from the measured results of the peak position shown in Fig. 9(b). It was estimated that there were less than 0.09° errors in the sum of θ₁ and θ₂ of the fabricated optical DMUX.

In order to evaluate manufacturing tolerances of our glass molding process, the length (L) of the manufactured optic block was measured for fifteen randomly selected samples. The measured values of L were between 6.655 mm and 6.679 mm. Although the measured values did not exactly match design value of 6.7 mm, manufacturing deviation was very small as ± 12 µm.

5. Conclusions

We have developed a highly alignment tolerant 4 × 25 Gb/s ROSA module for 100G Ethernet optical transceiver. To the best of our knowledge, the module has the largest lateral alignment tolerance of more than ± 250 µm among the 4 × 25 Gb/s ROSA, and this figure is more than 50 times larger than reported value of PLC type optical DMUX. The large alignment tolerance can be attributed to the dimensional tolerant optical DMUX which is based on the parallelogram-shaped optic block and four thin film filters.

We investigated dimensional tolerance for the optical DMUX block to define requirements for the length and angle of the optical DMUX block by using our three-dimensional simulation program. In addition, we verified the simulation results through geometric method with trigonometric functions. Finally, we measured the angular error value of the fabricated optical DMUX. According to the simulation results, the optical DMUX permitted length tolerance of ± 300 µm and angular tolerances of ± 0.1°. The fabricated optical DMUX was estimated to have the angular error of less than 0.09° between the mirror planes of the optic block. Since even a small error occurred in the fabrication process could be compensated by tilting the optical DMUX slightly, the ROSA module could be highly alignment and dimensional tolerant in alignment process.