Monolithic, Athermal Optical A/D Filter Vivek Raghunathan, Jurgen Michel and Lionel C. Kimerling Microphotonics Center, Massachusetts Institute of Technology, USA Collaborators: Prof. Karen K. Gleason, Dr. Jose Luis Yague, Dr. Jingjing Xu, ChemE, MIT Prof. Andrea Melloni, Stefano Grillanda, Politecnico Di Milano, Italy Prof. Kenneth B. Crozier, Shiyun Lin, Harvard School of Engg. 1
Outline 1. Motivation 2. Device: Passive athermal design 3. Material : Tailoring polymer properties 4. Process Integration 1. Multi-layer stacking: Hermetic sealing 2. Fabrication tolerance: Post-fabrication resonance trimming 5. System Design: Maximizing bandwidth 6. Concluding remarks 2
Motivation Demand for on-chip electronic-photonic synergy r Local temperature fluctuations : showstopper Athermal performance is critical to wavelength division multiplexing (WDM) integration r r Silicon-based WDM system with around 220 channels in the C band typical temperature fluctuations : 25 o C 125 o C r maximum allowable thermo-optic (TO) peak shift : less than 1 pmk -1 Thermal tuning forms a significant portion of NDD power in an ATAC environment (Stojanovic et al.) r 75% of energy cost at 1 GB/s per link Need for passive athermal design for low power operation 3
1. Optical components: Passive athermal device Desired TO resonance stability < 1 pm/k 4
Athermal prototype design Polymer Cladding TM Si SiO 2 Bottom Cladding Material a-si SiO 2 Si 3 N 4 Enablence Polymer (EP) n (@1550nm) 3.48 1.46 2.05 1.38 dn/dt ( 10-4 K -1 ) 2.3 0.1 0.4-2.65 Passive approach: Use negative TO effects from the clad material to compensate for the positive TO effects from waveguide core 5
Near complete TO compensation Key Observations: Slope : reflection of the waveguide dispersion Magnitude of the slope depends on waveguide aspect ratio TO coefficient contrast between the core and the cladding Tradeoff between athermal behavior and waveguide dispersion Athermal waveguide design varies depending on the wavelength to be filtered Vivek Raghunathan et al. Athermal Operation of Silicon waveguides: spectral, second order and footprint dependencies, Optics Express, vol.18, no. 17 (2010). 6
2. Optical component: Material selection Tailoring the optical properties of the polymer: Initiated chemical vapor deposition of p(pfda) and p(pfda-co-dvb) 7
Polymer design: Initiated Chemical Vapor Deposition (icvd) C-F bond has low index due to increased steric hindrance (high V) and small polarization (low R) n n 2 2 1 = + 2 R V R: molar refraction, V: molar volume 3 2 n dn = dx, where x= R 2 dt (1 x) dt V dn n n = dt 6n 2 2 ( 1)( + 2) α V Vivek Raghunathan et al. Co-polymer clad design for high performance athermal photonic circuits, Optics Express, vol.20, no.19 (2012) 8
Polymer design: Initiated Chemical Vapor Deposition (icvd) 2 2 dn ( n 1)( n + 2) = αv dt 6n a-si core: 550nm 205nm n (p(pfda)): 1.33 TO (p(pfda)): -2.1 10-4 n (p(pfda-co-dvb)): 1.38 TO (p(pfda-co-dvb)): -3.1 10-4 α V (DVB) : 8.6 10-4 α V (acrylate) : 0.5 0.9 10-4 Addition of DVB increases the index and TO coefficient by reducing the C-F bond density and increasing the volume expansion coefficient 100% improvement in FSR performance with the co-polymer cladding choice Vivek Raghunathan et al. Co-polymer clad design for high performance athermal photonic circuits, Optics Express, vol.20, no.19 (2012) 9
3. Optical components: Process integration CMOS BEOL integration: Multi-layer stacking Plasma stability Thermal stability Long term reliability 10
Dielectric Encapsulation: HDPCVD Need for Low temperature CVD process Conventional PECVD usually around 350 o C. High density plasma chemical vapor deposition using electron cyclotron resonance as the high density plasma source. Deposition at 130 o C. 500 nm of SiO 2 and SiN x Vivek Raghunathan et al. Stability of polymer-dielectric bi-layers for athermal silicon photonics, Optics Express, vol. 20, no.14, (2012) 11
Dielectric Encapsulation: TO performance 3% variation in thickness: 6 nm TO shift of 1 pm/k for every nm increase in thickness of film for TM mode. Athermal performance of the rings retained after dielectric encapsulation. Unaltered TO performance of hermetically sealed devices encourages multi-layer stacking and device fabrication Vivek Raghunathan et al. Stability of polymer-dielectric bi-layers for athermal silicon photonics, Optics Express, vol. 20, no.14, (2012) 12
4. Optical components: Fabrication tolerance Ability to trim the resonance within a 5 GHz (~ 42 pm) channel window 13
Photo-trimmable athermal resonators EP Leverage the visible light sensitivity of As 2 S 3 As 2 S 3 a-si 205 nm 500 nm SiO 2 Blue-shift of 10 nm (~1.2 THz) at a trimming rate of 8.5 pm/min (~1 GHz/min) Exposure to 10 mw/cm 2 of visible light (450-650 nm) Δn eff : -0.024, Δn As2S3 : -0.13 TO shift of < 1pm/K and trimming rate of 1 GHz/min satisfies the requirements of Si WDM system with a 20 GHz channel spacing Vivek Raghunathan et al. Trimming of athermal silicon resonators, Advanced Photonics Congress, Technical digest (CD) OSA (2012), paper IW4C.5 14
5. System design: Optimization and tradeoff Maximizing aggregate bandwidth of an add-drop filter Channel count performance Power handling capacity 15
Channel design of an add-drop filter High Q: smaller channel spacing, High FSR/Small footprint: large channel count Vivek Raghunathan et. al. Add-drop channel counts limitation by athermal filter design, Manuscript under preparation. 16
Channel counts performance opt opt α ( R ) ( CR 1) = α b b b sc α sc dominated regime (α sc > α b ) r Increasing FSR increases the channel count for a given α sc of 2 db/cm α b dominated regime (α sc < α b ) r Increasing FSR is offset by increased channel spacing for a given cross-talk thereby decreasing the channel count Existence of an optimum FSR that maximizes the channel count Vivek Raghunathan et. al. Add-drop channel counts limitation by athermal filter design, Manuscript under preparation. 17
Power handling capacity Decreasing effective TO Red-shifts at high launch powers are due to thermal effects induced by TPA/FCA For no appreciable variation in resonance (< 10 pm), the power handling capacity increases 10X when going from SOI waveguide (~ 12 dbm/ 15.85 mw) to athermal waveguide (~ 22 dbm/ 158.5 mw) 10x improvement in the power handling capacity of passive athermal design Vivek Raghunathan et. al., High capacity photo-trimmable athermal silicon waveguides, IEEE International Conference on Group IV photonics (2012). 18
Aggregate bandwidth performance Receiver sensitivity: 100 nw/gb/s Max power of highly confined SOI : 15.8 mw Max power of athermal SOI: 158 mw Athermal design outperforms conventional rings in the aggregate bandwidth capability due to its superior power handling capacity Vivek Raghunathan et. al. Add-drop channel counts limitation by athermal filter design, Manuscript under preparation. 19
Aggregate bandwidth performance Receiver sensitivity: 100 nw/gb/s Max power of highly confined SOI : 15.8 mw Max power of athermal SOI: 158 mw Athermal design outperforms conventional rings in the aggregate bandwidth capability due to its superior power handling capacity Vivek Raghunathan et. al. Add-drop channel counts limitation by athermal filter design, Manuscript under preparation. 20
Summary Lowest reported TDWS value of 0.5pm/K for Si resonators. Athermal filter waveguide width depends on the wavelength to be filtered out. Optical properties of an acrylate polymer can be tailored by changing its C-F bond density with the right cross-linker Hermetic sealing using dielectric cap encourages the possibility of multi-layer stacking on polymer cladded devices. Thin photosensitive layer of As 2 S 3 enables visible light trimming for fabrication tolerances. High capacity of athermal rings outweighs its low FSR thereby enabling higher aggregate bandwidth compared to conventional rings 21
Acknowledgment We would like to acknowledge Tomoyuki Izuhara from Enablence for help with the polymer coating. We would like to thank Dr. Anu Agarwal and Vivek Singh from EMAT group for their contribution to the trimming project This thesis work has been sponsored under the Defense Advanced Research Projects Agency s Athermal Photonic Circuits (APhoCs) program, Fully LASER integrated photonics program under APIC corporation and DARPA UHPC program 22
THANK YOU!! QUESTIONS? 23
Introduction Si based photonic interconnects r Bandwidth, density and latency advantages over electrical interconnects [1,2] r More power efficient than electrical links at high data rates [1] Si based Wavelength Division Multiplexing (WDM) r Integral part of optical communication r r Bandwidth scaling : Tera-scale computing Ring resonator based filters Photonic CMOS based WDM [1] 1. Young et al. Optical I/O Technology for Tera-Scale Computing, IEEE Journal of Solid-State Circuits, Vol 45, No. 1, Jan 2010 2. Krishnamoorthy et. al. Computer systems based on Silicon Photonic Interconnects Invited Paper, Proceedfings of IEEE, Vol.97, No.7, July 2009 24
Prototype design Design Parameters dnc dt dncl,, Γ( nc, ncl, geometry, λ) dt Core dimensions: 700nm 206nm r Single mode TM operation TOP VIEW of unclad a-si ring and bus Material a-si SiO 2 Si 3 N 4 Enablence Polymer (EP) n (@1550nm) 3.48 1.46 2.05 1.3793 dn/dt ( 10-4 K -1 ) 2.3 0.1 0.4-2.65 25
Polymer design: Initiated Chemical Vapor Deposition (icvd)* icvd of p(pfda) and p(pfda-co-dvb) to study the effect of cross-linking on polymer optical properties *Kenneth et al. Initiated Chemical Vapor deposition of Poly (alkyl acrylates): An Experimental Study, Macromolecules 2006, 39, 3688-3694 *Kenneth et al. Initiated Chemical Vapor deposition of Poly (alkyl acrylates): A Kinetic model, Macromolecules 2006, 39, 3695-3703 26
Polymer design: Initiated Chemical Vapor Deposition (icvd) Athermal cross-sections ppfda : 550 nm 190 nm p(pfda-co-dvb): 550 nm 212 nm Bending Q of 10 4 ppfda : R b : 12.5 µm, FSR: 10 nm p(pfda-co-dvb): R b : 5 µm, FSR: 20 nm 100% improvement in FSR performance with the co-polymer cladding choice Vivek Raghunathan et al. Co-polymer clad design for high performance athermal photonic circuits, Optics Express, vol.20, no.19 (2012) 27
Polymer BEOL Integration: Plasma and Thermal Stability Confidential Need for low temperature deposition of dielectric to retain polymer functionality Need for dielectric encapsulation for robust performance in oxidizing atmosphere and to prevent polymer degradation in plasma Vivek Raghunathan et al. Stability of polymer-dielectric bi-layers for athermal silicon photonics, Optics Express, vol. 20, no.14, (2012) 28
Dielectric Encapsulation: HDPCVD 29
Polymer BEOL Integration: UV stability Confidential UV curing of polymer cladding after deposition using short arc mercury lamp 5 mw/cm 2 for 20 minutes : 6000 mj/cm 2 Baking at 150 o C under vacuum for 20 minutes FTIR confirms no bond chemistry change on further UV exposure. Unaltered TO performance on further UV exposure (9.5 mj/cm 2 ). UV curing and baking ensures robust polymer performance during multi-layer stacking Vivek Raghunathan et al. Stability of polymer-dielectric bi-layers for athermal silicon photonics, Optics Express, vol. 20, no.14, (2012) 30
Footprint constraints of athermal design Material n Γ n eff n g R b (µm) Si 3.48 0.58 1.78 3.68 3.5 Si 3 N 4 2.05 0.9 1.75 2.2 5.5 For a given bending loss of 4.34 db/cm r Comparable R b under athermal constraint Δn/TO : good indicator for core material selection. LIC performance comparable to HIC systems under athermal constraint Vivek Raghunathan et al. Athermal Operation of Silicon waveguides: spectral, second order and footprint dependencies, Optics Express, Vol.18, 2010, 17631-17639 31
Dependence on scattering loss For a channel count of 256 α sc (channel TM) < 2.12 db/cm α sc (channel TE) < 2.62 db/cm α sc (slotted TM) < 5.05 db/cm α sc (slotted TE) < 6.08 db/cm TM has lower scattering loss than TE due to lesser overlap with the sidewall Athermal slotted cross-section has a higher loss threshold Optimum mode selection (TE vs. TM) is driven by the fabrication technique that determines the sidewall roughness Vivek Raghunathan et. al. Add-drop channel counts limitation by athermal filter design, Manuscript under preparation. 32
Bending loss performance @ 1550 nm Slotted cross-sections r Higher n eff at athermal: lower expansion into the BOX r Mode-pinning in the slot : lower bending loss Athermal slotted cross-sections have superior bending loss and FSR performance compared to the channel counterparts Vivek Raghunathan et. al. Add-drop channel counts limitation by athermal filter design, Manuscript under preparation. 33
(a) l 1, l 2 IN DROP Through port Cross-talk THROUGH l 1 l 2 (a) Intensity [db] 0-5 -10-15 -20 +8 dbm -25-8 dbm -30 1560.8 1561 1561.2 1561.4 1561.6
Through port (with thermal compensation) (a) Intensity Intensity [a.u.] [db] 0-5 -10-15 -20-25 +8 dbm compensated -8 dbm +8 dbm Drop transmission - 8 dbm -30 1560.8 1561 1561.2 1561.4 1561.6 Wavelength [nm]