Intra- and Inter-Wafer Characterization of Waveguide Propagation Loss and Reflectivity Mohammad Istiaque Reja,1,2,* 1

Dept. of Electrical and Electronic Engineering, Chittagong University of Engineering and Technology, Bangladesh 2 Institute of Communication, Information and Perception Technologies, Scuola Superiore Sant’Anna, Pisa, Italy * [email protected]

Abstract—In silicon photonics, due to the increasing complexity of the device designs and the increasing density of the functions integrated on a single circuit, the precise characterization of the building blocks has become essential in order to assess the quality of the fabrication outcomes. As waveguides are the fundamental building block of photonic integrated circuits, its accurate characterization is very important to evaluate the quality of all passive elements. In this paper the characterization of two very important parameters, namely waveguide propagation loss and waveguide sidewall reflectivity, are reported for waveguides utilized inside a microring-based silicon photonic network-on-chip. Using a recently proposed method based on undercoupled all-pass microring structure, the two parameters from five differently positioned chips in two different wafers are measured in order to investigate the intra- and inter-wafer variations. Precise measurements allow to define range of values and variations for these two parameters in photonic network-on-chip. Index Terms—Waveguide propagation loss, sidewall reflectivity, network-on-chip (NoC), silicon photonics.

I. INTRODUCTION To increase the computational performance in an energy efficient way, parallelization is currently adopted by placing multiple cores processors on a single chip [1]. The on-chip interconnection network or network-on-chip (NoC), providing communication between the cores and between the core and cache in the multi core architecture, must fulfil the high bandwidth and low latency requirements in a power efficient manner [2]-[7]. Due to the limitation of conventional electrical NoC, photonic solutions have been researched in recent times, aimed at providing high throughput and minimal access latencies along with very low power consumption independent of the capacity [8]. Silicon-on-insulator (SOI) platform is the best choice for the next generation of photonic NoC, because SOI platform can provide high refractive index difference thus allowing to make circuits with small footprint. Also, silicon is CMOS compatible and it can be used to make optoelectronic chip in a cost effective way [9]. Finally, in the recent years the maturity has improved significantly, allowing to demonstrate a variety of high-performance integrated devices [10]. With the advancement of the technology not only the designs of the device are becoming more and more complex but also the number of functions integrated onto a single circuit is increasing at the same pace [11]. So, it has become more important than before to characterize the building blocks precisely in order to know the discrepancies between the fabrication outcomes and the intended device. Precise characterization is very important, because it allows to

understand the tolerances which in turn allow to improve further design cycles [12]. The propagation loss and the reflectivity arising from sidewall roughness of the silicon waveguides are two important parameters in integrated optical devices, which impact the performance of the whole silicon photonic circuit [13]. So, the characterization of the waveguide loss and reflectivity not only help us to know the quality of the waveguide in terms of fabrication accuracy but also gives us a hint about the quality of the other passive devices on the Silicon chip. Most of the propagation losses in Silicon nanowire arise mainly due to the sidewall roughness and this roughness is a result of the fabrication process, more specifically lithography and etch process [13]. Many methods have been developed over the last three decades for measuring the loss accurately [14]-[17]. The most commonly used method is 'cut back' method [15]. This method, although provides fast and direct measurement, has limited accuracy, because it is very difficult to maintain the same coupling efficiency for every measurement [16]. In the sliding prism method, it is required to extract the light with 100% efficiency. Moreover, quite complicated mechanical arrangement is needed [17]. The accuracy of the conventional Fabry-Perot (FP) resonance method is limited by the requirement of the exact estimation of the facet reflectivity. The experimental procedure is also very difficult, since it is essential to employ narrow line width stabilized source [14]. The modified FP method used the curve fitting approach to simplify the experimental procedure and allows to use broadband low-coherence source rather than narrow line width high coherence source. Also, the facet reflectivity and loss can be measured simultaneously without assuming either of the two [14]. A new method based on undercoupled all-pass microring structure is proposed in [10], which measures the waveguide loss and sidewall reflectivity simultaneously through curve fitting. This method is independent of input coupling losses and it is tolerant to the bus to ring gap choice, thus measures the loss very accurately. The concept of this method is based on the interplay between standing wave modes and travelling ring modes. In this paper, the characterization of the propagation loss and sidewall reflectivity of five differently positioned chips from two wafers using this method will be reported, which allows to investigate the intra- and inter-wafer variations of these two parameters. As this method measures the loss and reflectivity very precisely, the measurements will allow to define range of values and variations of the loss and

reflectivity in photonic NoC and will be used to assess the basic switching element of multimicrorinng network in two different types of architectures: doped slabb case and undoped slab case. II. LOSS AND REFLECTIVITY CHARAC CTERIZATION A. Brief description of the model The loss measurement method based onn undercoupled allpass microring structure [13] will be useed in this paper to characterize the intra and inter wafer losss and reflectivity. Unlike the basic model found in the litterature [18], [19], where it is considered that a single unidirecctional mode of the resonator is excited, this model considers the propagation in t bus in order to the both directions both in the ring and the take into account the backward proopagation due to reflections . The Transmission (T) at 'through' portt of the bus when only 'input' port is excited can be written ass [13]: (1 ) 1 where 1 In this equation, t is the field transmisssion coefficient, a and are the field loss and accumulation of phase over one complete round trip respectively, Rc is the reflection coefficient in both directions which arises a due to the scattering caused by sidewall roughness. . where, r is the radius of the a is expressed as, ring and α is the energy propagation loss per unit length in , where R is the dB/cm. Rc is expressed as, reflectivity which is a stochastic average annd is the phase of reflection coefficient which is approxximated, assuming lossless circuits, by the equation [20] . Through fitting the simulated curve froom this model with the experimentally measured spectra, thiss method estimates the waveguide propagation loss and reflectivity simultaneously in undercoupled conditionn. As undercoupled spectrum responses are highly sensitive too these parameters, this method allows to measure loss andd reflectivity very accurately.

C - Center chip TC - Top Corner chip

B. Chip characterization detaills Chips were manufacturedd by IME through CMC microsystem in a multiprojecct full-active wafer run. For characterization, five differentlly positioned chips from two different wafers are selected. The positions are chosen in order to cover almost all the reggions of the wafer. One chip is chosen from the center and the other four chips are from four corners as shown in Fig.1. Therefore the measurementss and the analysis of all these differently positioned chips will w allow to define range of values and variations of the losss and reflectivity across all the positions of two wafers. C. Device architecture and dessign A specific structure compossed of a bus of all-pass rings [13], is used in order to measuure the waveguide propagation loss. In Fig. 2, the mask layout of o the used structure is shown. It can be noticed that the sam me structure is present twice. Actually the upper one is chaaracterized by a 90-nm thick internal slab of doped Si in the t microring whereas in the lower one the same slab is undooped. This internal doped slab enabled the thermal tuning of the microrings of the characterized Network-on-Chipp (NoC) independently [7], [21]. The ring with the doped region is designed to minimize the extra loss due to the dopingg. In this paper, it will be also investigated whether this dopedd region has induced any extra waveguide losses in comparisonn to the undoped one. In this loss measurement scheme six microrings are coupled with a bus waveguidde as shown in Fig. 2. The distance of the microrings from m each other is large enough (about 39.5 micron) so that thhey do not interact with each other. By varying the gap betw ween the bus and the ring the amount of coupling is varied in i each of the rings. The left most ring has a gap of 200 nm from f the bus and increasing at a step of 50 nm the right most ring r has a gap of 450 nm. The gaps are chosen in order to havve some rings in overcoupled and some rings in undercoupledd conditions. The radii of the rings are in the t order of 10 µm in order to have a negligible bending loss in i comparison to the scattering loss. To distinguish the resonnances of the six rings it is necessary to make their free spectral range different from each other. So, the radius has been b changed slightly in all of the rings. Starting from the left most ring the radius has been increased at a step of 25 nm. As a consequence the free spectral range in each ring will differ by 1 nm approximately [13].

200 nm gap

250 nm gap

300 nm m gap

350 nm gap

400 nm gap

450 nm gap

LC - Left Corner chip RC - Right Corner chip BC - Bottom Corner chip

(a) Doped slaab structure

(a) Undoped slab structure Fig. 1 The position of the characterized chipps on the wafer Fig. 2 Mask layout of the used structuure for waveguide loss measurement

(a)

(b) Fig. 3 Device cross-section scheme in the coupling reegion of (a) the structure where doped slab is used in the ring [referring to Fig. 2 (a)] (b) the structure where undoped slab is used in the ring [referring to Fiig. 2 (b)]

The cross-section of the bus and ring in the coupling region are schematically shown in Fig. 3.. The cross-section of the bus consists of a single-mode ridge waveguide having a width of 460 nm and a height of 220 nm, whereas the crosssection of the ring consists of a sinngle-mode half-rib waveguide having a width of 480 nm and 220 2 nm height. Fig. 3 (a) shows the 90 nm thick internal slabb of doped silicon 1µm away from the microring waveguide whereas Fig. 3 (b) shows fully undoped internal slab, whicch are used in the upper and lower structure of Fig. 2 respectiively.

E. Results and discussions By varying the loss (α), reflectivity (R) and field transmission coefficient (t) the simulated curve is fitted with the undercoupled resonances off the experimentally measured spectrum. In this way the loss and reflectivity from both the doped and undoped architectuure of differently positioned chips of two wafers are estimateed. In Fig. 5, a histogram showinng the statistics of losses from 11 samples of doped slab structure and 11 samples of undoped slab structure is preesented. For the doped slab structure the average loss is 2.62 2 dB/cm and the standard deviation is 0.30, whereas for the t undoped slab structure the average loss is 2.14 dB/cm and the standard deviation is 0.55. It can be noticed from the Fig. 5 that in the doped slab architecture a slightly larger loss occurs than in the undoped case (with the entire distributioon shifting toward higher loss) due to the loss contribution of the t field tails propagating into the doped slab. In Fig. 6 the loss variatioons across two wafers from differently positioned chips are provided. Fig. 6 (a) and Fig. 6 (b) show the variation for the undoped slab and doped slab case respectively. Also the maximum, m minimum, average value of loss are provided by thee upper, lower and middle line respectively. This figure gives us a complete description of all the loss information for thesee two wafers. It can be noticed that the average loss in both thhe wafers are almost equal for both the doped and undoped struuctures. doped slab structure

undoped slab structure

8

Fig. 4 Experimental setup for spectrum measurement m

3.2 3

2.8 2.6 2.4 2.2 2 1.8 1.6 1.4

S.D. = 0.56

1.2

S.D. = 0.53

Waveguide propagation loss [dB/cm]

Waveguide propagation loss [dB/cm]

Frequency

D. Experimental measurement 6 During the experimental measurement, spectral s scans have 4 been obtained by generating a continuous wave signal using an external cavity tunable laser (TL) in thhe range of 1500 1600 nm with a power of 0 dBm. After that, t a polarization 2 controller (PC) maximizes the coupling efficiency e between the fiber and the input grating coupler of thhe device under test 0 (DUT). A synchronized power meter is used u to collect the 0-0.5 0.51-1 1.01-1.5 5 1.51-2 2.01-2.5 2.51-3 3.01-3.5 output spectrum from the 'through' port grating coupler. All Propaggation Loss (dB/cm) spectra have been collected with a spectral resolution of 1 pm. The measurements have been carried out o maintaining a constant temperature of 25°C by usinng a temperature Fig. 5 Statistics of waveguide propagaation losses measured from 22 samples controller. In Fig. 4 the experimental setuup for the spectral (11 each from doped and undoped slab structures) measurements is schematically shown. Wafer1 Wafer2 Wafer1 Wafer2 Spectral scans have been taken from booth the doped slab and undoped slab structures. These measurements m are performed for all the differently positioneed chips (as shown in Fig. 1) fabricated from two wafers. 3.2 3

2.8 2.6 2.4 2.2 2 1.8 1.6 1.4 1.2

S.D. = 0.20

S.D. = 0.38

Undoped slab architecture

Doped slab architecture

(a)

(b)

Fig. 6 Variation of the waveguide propagation losses across two wafers from differently positioned chips measured using (a) undoped slab architecture and (b) doped slab architecture. For eaach dataset the upper line denotes the maximum, lower line denotes the minim mum, and middle blue line denotes the average propagation loss. Also for eachh dataset the Standard Deviation (S.D.) has been provided

(a) Wafer 1 undoped structure

(b) Wafer 2 undoped structure

Therefore the utilized method is shown to measure the loss very accurately since it is independent of input coupling losses and reflections and works on undercoupled rings, where losses are dominant. Fig. 8 shows the reflectivity variations across two wafers from differently positioned chips for both the doped and undoped case. It can be seen that for both types of architecture the maximum reflectivity value is of the order of 0.01-0.012 and minimum reflectivity value is of the order of 0.0025-0.004. The average value of reflectivity lies in the range of 0.006-0.007 for both wafers. Fig. 9 shows the intra-wafer reflectivity distributions in two wafers at differently positioned chips for both undoped and doped structures. No specific pattern for the reflectivity is found: in particular no dependence on the position of the chip in the wafer or on the chosen wafer is highlighted, nor any dependence on the presence or absence of the doped slab, as expected. Wafer1

Fig. 7 Intra-wafer loss distributions in both wafers for undoped (a, b) and doped case (c, d)

In Fig. 7 the intra-wafer loss distributions in two wafers at differently positioned chips for both undoped and doped structures are shown. It can be seen, except for wafer 2 doped structure, that the waveguide propagation loss values are typically lower at the center due to the more accurate processing in the center of the wafer, thanks to better fabrication process outcome. In table 1, a comparison is presented between the loss characterized in this paper and loss characterized by the foundry through their cutback PCM structures at wavelengths in the range 1530 - 1580 nm for undoped channel and rib waveguides [22]. It can be seen from the table that the losses measured in this paper are in line with the data of the foundry. The foundry measured the loss for channel straight waveguides and rib straight waveguides (i.e. slabs are on both sides of the waveguide), but in our case we measured the loss for half-rib 10-µm radius curved waveguide, large enough to avoid the additional effect of the bending loss to the total propagation losses. As expected to fully validate the method, the loss measured here and reported in table 1 is in the middle between the channel waveguide and rib waveguide, not far from rib loss values, measured by foundry.

Wafer1 0.014

0.012

0.012

0.01

0.01

Reflectivity

(d) Wafer 2 doped structure Reflectivity

(c) Wafer 1 doped structure

Wafer2

0.014

0.008

0.006

0.004

0.008

0.006

0.004

0.002

0

Wafer2

0.002

S.D. = 0.0035

S.D. = 0.0031

Undoped slab architecture

(a)

S.D. = 0.0027

S.D. = 0.0021

0

Doped slab architecture

(b)

Fig. 8 Variation of the reflectivity across two wafers from differently positioned chips measured using (a) undoped slab architecture and (b) doped slab architecture. For each dataset the upper line denotes the maximum, lower line denotes the minimum, and middle blue line denotes the average reflectivity. Also for each dataset the Standard Deviation (S.D.) has been provided

(a) Wafer 1 undoped structure

(b) Wafer 2 undoped structure

(c) Wafer 1 doped structure

(d) Wafer 2 doped structure

TABLE I COMPARISON BETWEEN LOSS CHARACTERIZATION BY THE FOUNDRY AND LOSS CHARACTERIZATION IN THIS PAPER FOR UNDOPED STRUCTURE

Characterization by foundry [22] Used method and structure

Cutback PCM structures

Waveguide width

500 nm

Type of waveguide Loss (dB/cm)

Channel waveguide 1.75

Characterization in this paper All pass filter structure in undercoupled condition 480 nm

Rib waveguide 2.2

Half-rib waveguide (slab only in the ring internal side) Wafer 1 Wafer 2 2.13 2.15

Fig. 9 Intra-wafer reflectivity distributions in both wafers for undoped (a, b) and doped case (c, d)

III. CONCLUSION This paper presents the characterization of two very important parameters of next generation of Si photonic NoC, namely: the waveguide propagation loss and the sidewall reflectivity. Measurements from five differently positioned chips of two wafers show the intra and inter wafer variations of these two parameters. The waveguide propagation loss is usually lower in the center of the wafer due to more accurate processing in this region. Quite low value of average loss is found in the waveguide of this recently developed all-optical photonic NoC and it is around 2.1 dB/cm. No specific intrawafer pattern is found for the value of reflectivity. The maximum reflectivity value is of the order of 0.01-0.012 and minimum reflectivity value is of the order of 0.0025-0.004. The average value of reflectivity lies in the range of 0.0060.007. ACKNOWLEDGMENT The author would like to thank the Institute of Microelectronics (IME), Singapore for foundry service. The author would also like to thank the CMC Microsystems for providing the characterization data. Special thanks to Dr. Nicola Andriolli (Assistant Professor, SSSUP, Italy) and Fabrizio Gambini (Researcher, CNIT-National Laboratory of Photonic Networks, Pisa, Italy) for valuable suggestions and feedbacks throughout this work. REFERENCES [1]

[2]

[3] [4] [5] [6]

C. Li, M. Browning, P. V. Gratz, S. Palermo "LumiNOC : A Power-Efficient, High-Performance, Photonic Network-on-Chip", IEEE Transactions on Computer Aided Design of Integrated Circuits and Systems, vol.33, no.6, June 2014 Batten, C., Joshi, A., Stojanovć, V., & Asanović, K. "Designing chiplevel nanophotonic interconnection networks", Integrated Optical Interconnect Architectures for Embedded Systems, pp. 81-135,. Springer New York, Sep, 2012. J. Psota et al., “ATAC: improving performance and programmability with on-chip optical networks,” in Proc. of IEEE International Symposium on Circuits and Systems (ISCAS), 2010, pp. 3325 - 3328 . P. Pintus, P. Contu, P. Raponi, I. Cerutti, and N. Andriolli, “Siliconbased all-optical multi microring network-on-chip” Opt. Lett., vol. 39, no. 4, pp. 797-800, 2014. S. Faralli, F. Gambini, P. Pintus, I. Cerutti, and N. Andriolli, “Ring versus bus: A BER comparison of photonic integrated networks-onchip,” in Proc. IEEE OI 2015, pp.22-23. Andriolli, N.; Cerutti, I.; Pintus, P.; Scaffardi, M.; Marini, D.; Montanari, G.B.; Mancarella, F.; Ferri, M.; Balboni, R.; Bolognini, G., "Challenges and progress toward a silicon-based multi-microring optical network-on-chip," in Proc. EUCNC 2014, pp. 1-5.

[7]

[8]

[9] [10]

[11] [12]

[13]

[14] [15] [16]

[17] [18] [19] [20] [21]

[22]

F. Gambini, S. Faralli , P. Pintus, N. Andriolli, and I. Cerutti "BER evaluation of a low-crosstalk silicon integrated multi-microring network-on-chip" Optics Express, Vol. 23,Issue 13, pp. 17169-17178, 2015 A. Shacham, K. Bergman, L.P. Carloni “Maximizing gflops-per-watt: high-bandwidth,low power photonic on-chip networks” Proc. Third Watson Conf. Interaction between Architecture, Circuits, and Compilers, pp.12–21, 2006 H. Ting et al. "Silicon photonic network-on-chip and enabling components", Science China Technological Sciences March 2013, vol. 56, issue 3, pp 543-553 Y. Li, Y. Zhang, L. Zhang, and A. W. Poon, "Silicon and hybrid silicon photonic devices for intra-datacenter applications: state of the art and perspectives [Invited]," Photon. Res. vol. 3, issue 5, pp. B10B27, 2015 Kuramochi et al. "Large-scale integration of wavelength-addressable all-optical memories on a photonic crystal chip". Nature Photon. 8, 474–481(2014). R. Bruck, B. Mills, B. Troia, D. J. Thomson, F. Y. Gardes, Y. Hu, G. Z. Mashanovich ,V. M. N. Passaro, G. T. Reed and O. L. Muskens, "Device-level characterization of the flow of light in integrated photonic circuits using ultrafast photo modulation spectroscopy" Nature Photonics 9, 54–60 (2015) M. Moresco, M. Romagnoli, S. Boscolo, M. Midrio, M. Cherchi, E. S. Hosseini, Douglas Coolbaugh, Michael R. Watts, and Birendra Dutt, "Method for characterization of Si waveguide propagation loss" Optics Express, Vol. 21,Issue 5, pp. 5391-5400(2013) S. Taebi, M. Khorasaninejad, and S. S. Saini, “Modified Fabry-Perot interferometric method for waveguide loss measurement” Appl. Opt. 47(35), 6625–6630 (2008). A. Rickman, G. T. Reed, B. L. Weiss, and F. Namavar, "Low-Loss Planar Optical Waveguides Fabricated in SIMOX Material", IEEE photonics technology letters, vol. 4, no. 6, June 1992. S. Chen, Q. Yan, Q. Xu, Z. Fan, J. Liu, "Optical waveguide propagation loss measurement using multiple reflections method," Elsevier Optics Communications, vol. 256, no.1–3, pp. 68–72, December 2005. J. Wei et al., "Improved method for loss measurements in planar waveguides," Elsevier Optics and Lasers in Engineering vol. 45, pp. 419–422, 2007. A. Yariv, “Universal relations for coupling of optical power between microresonators and dielectric waveguides”, Electron. Lett. 36(4), 321–322 (2000). J. Heebner, R. Grover, T. Ibrahim, "Optical microresonator: theory, fabrication and application", Springer series in Optical Sciences, ISBN: 978-0-387-73068-4 R. E. Collin, "Foundation for Microwave Engineering", 2nd ed. , McGraw-Hill, N.Y., 2000. M. R. Watts, W. A. Zortman, D. C. Trotter, G. N. Nielson, D. L. Luck, and R. W. Young, “Adiabatic resonant microrings (ARMs) with directly integrated thermal microphotonics,” in Proc. CLEO/QELS 2009. Measurements data from Institute of Micro Electronics (IME) provided by CMC Microsystems.