Compact, low capacitance optical modulators are vital for efficient, high-speed chip to chip optical interconnects. Electro-optic (EO) polymer cladding micro-ring resonator modulators have been fabricated and their performance is characterized. Optical modulators with ring diameters smaller than 50 µm have been demonstrated in a silicon nitride based waveguide system on silicon oxide with a top cladding of an electro-optic polymer. Optical modulation has been observed with clock signals up to 10 GHz.
©2008 Optical Society of America
As data rates increase, it is becoming more challenging to transmit electrical signals with good signal integrity and low power consumption. Chip to chip communication such as CPU to memory, CPU to graphics chip, or CPU to CPU are just a few near term areas where increasing data rates give rise to performance bottlenecks. Eventually long term challenges can occur in meeting the bandwidth requirements and minimizing the delay within the CPU for core to core communications in many core processors. Optical interconnects are one possible solution to these problems.
A practical optical interconnect for insertion into these applications must possess high data rates, large bandwidths and low power. One approach to power efficiency is miniaturizing the components such that the capacitance is very low. Miniaturization also allows for dense packing of the components which keeps cost down by keeping die size as small as possible.  Some form of multiplexing, such as wavelength division multiplexing (WDM) to minimize the number of fibers or waveguides to complete the link for chip to chip applications, may be needed. 
Our goal is to make optical components for an optical link that is not only compatible with advanced CMOS processing, but can be fabricated monolithically with transistors without sacrificing transistor performance. Although CMOS compatible photonics has been demonstrated, the substrate and thermal budget were not compatible with advanced CMOS in the sub-45nm technology nodes; a major drawback for future CPU applications. A schematic of our approach is seen in Fig. 1. We call this approach “photonics on top of CMOS” or “photo-CMOS” because we are creating a monolithic optical layer directly on top of the existing CMOS chip.
A fiber coupled external CW laser is used as the optical power supply. This laser could be multi-wavelength laser or a bank of multiple lasers MUXed into a single fiber. The fiber is coupled to an on-die waveguide. The light can then be divided and sent to several modulator arrays (not shown). The ring resonator modulators are wavelength specific and modulate only a single laser frequency. CMOS circuitry on the die drives the modulators and converts electrical signals into optical signals. The light signals are then sent off-die into fibers (or board level waveguides) where they are then coupled into die level waveguides. Finally the light signals are coupled through deMUXing ring resonators and coupled into photodetectors. The photodetectors covert the light signals into electrical current which is amplified and converted into voltage swings using a transimpedence amplifier in the CMOS circuitry.
This approach allows integration of the optical components into the back-end interconnect layers of the die. The fabrication of the optical components is decoupled from the electrically active components. This is desirable for several reasons. First of all, since the optics reside on a separate layer from the transistors, the die area can be conserved as the optical layer can be placed directly on top of the transistor layer. Secondly, the optical layer can be readily added to different process generations or technologies. For instance, the optics can be added to the CPU and Chipset or memory made on different technology nodes. The main drawback to photo-CMOS, is that they must be fabricated at a lower temperature process ceiling (~450 C) imposed by back end of the line processing. Our approach to keeping within the thermal budget of the back end of the line process is to use an EO polymer based ring resonator optical modulator as the electrical to optical converter and a waveguide coupled Ge on oxide metal semiconductor metal (MSM) photodetector as the optical to electrical converter.
An EO polymer based ring resonator modulator has many desirable characteristics. First of all, the electro-optic coefficients of EO polymers have been reported at >300 pm/V  or >10 times larger than that of lithium niobate, the industry standard modulator material for the telecom industry. Next, the mechanism for the high electro-optic effect is a shuttling of the electrons within the molecular orbital of the chromophores which is inherently very fast and has been employed in the demonstration of an optical modulator >100 GHz.  The dielectric constant of the EO polymers are very small ε~2.5–4 , leading to very low capacitances. In conjunction with a ring resonator modulator, low capacitance EO polymers allow high speed devices that act as a simple lumped element capacitor. This avoids the need for low loss traveling wave electrodes which consume considerable power due to the typical 50 Ω termination on the electrodes. Finally, the EO polymers can be spin cast, such that the processing temperatures for post-spin anneals and poling can be considerably lower than the 450°C process ceiling. Since a ring resonator modulator is wavelength selective, it can be exploited to making a WDM interconnect system without the need of additional DEMUX.
In this paper we will focus on the EO polymer based optical modulator. The photodetector is a waveguide coupled Ge on SiO2 MSM photodetector and has been described elsewhere. 
2. Modulator design and fabrication
Conventional waveguide designs for EO polymers consist of a ridge waveguide fabricated out of the EO polymer itself, surrounded by top and bottom polymer cladding.  This approach generally leads to a low index contrast between the waveguide and cladding (often <0.2). Since the refractive index of EO polymers is typically within the range, n ~1.5–1.7, this leads to relatively large optical mode sizes and prevents the controlling electrodes from being spaced closely together . It also makes for relatively large minimum bend radii of the waveguides, thereby limiting miniaturization of the modulators.  The smallest reported ring resonator modulator based on EO polymer waveguide to date has a 300 µm radius.  Typical EO polymer modulators have ~10 µm electrode spacing. To help shrink the device dimensions, we chose to use the EO polymer as a top cladding material and allow for higher index contrast waveguide systems to give more compact devices and electrode spacing. The core waveguide in our modulator is 450 nm thick silicon nitride (n=2) deposited by PECVD using a deposition temperature of 400 °C. The silicon nitride waveguides are patterned with a plasma etch using conventional optical lithography in various widths ranging from 0.5–0.9 µm. The linear EO effect is used to modify the index of the EO polymer which modifies the optical mode effective index and imparts an index change that modulates the intensity of light. Conventional Cu damascene co-planar electrodes are used to apply the controlling electric field.
The EO polymer (AJTB141/APC) is a guest-host polymer made by doping 28 wt% of AJTB141chromophore into an amorphous polycarbonate (APC)) (Fig. 2).
The solution of this polymer (11–12 wt % of concentration) was filtered through a 0.2 µm PTFE-syringe filter and spin-coated over the nitride waveguide and Cu electrodes. The resulting films were baked in a vacuum oven at 85 °C overnight to ensure the removal of any residual solvent, and the film thickness was 0.7 µm measured by profilometry. The optimal poling temperature is at 135 °C, around the glass transition temperature of the polymer with a field of 80–100 V/µm.
3. Results and discussion
Ring resonator based modulators offer the advantage of very small size, compared to Mach-Zehnder interferometer modulators. These modulators have been demonstrated with several materials systems, including Si and EO polymers. By tuning the index of refraction of the EO polymer, the resonance condition is changed, causing a shift in the resonance location. When a fixed wavelength of light is used in the waveguide, the shift in resonance can be used to turn on and off the light. (see Fig. 2).
With this system, we have achieved ring resonators as small as 21 microns in radius with electrodes spaced at 3 microns. The 21 µm radius resonator has a free spectral range (FSR) of 7 nm. The resonators have exhibited Q values as high as 40,000 and extinction depths of > 15 dB. Higher Q and deeper extinction result in larger modulation depth at the same drive voltage. A high Q resonator, on the other hand, could limit the modulator bandwidth; therefore, a trade-off between the bandwidth and sensitivity of the modulator must be taken into account. Q’s between 5,000 and 10,000 would have sufficient bandwidth for 40 Gb/s modulation.
To make these ring resonators (Fig. 3(a)) optically active, we have selectively etched away the top oxide cladding (Fig. 3(b)) and have replaced it with electro-optically active polymer previously described. Since the EO polymer has a higher index (n~1.61) than the silicon oxide bottom cladding (n~1.48), the system is now considered to be asymmetric and careful design of the waveguide dimensions is needed to optimize modulator performance.
The electro-optic polymer’s index of refraction changes with the application of an electric field described by the following equation:
Where Δnpolymer is the change in index of refraction of the polymer, n is the index of the polymer, reff is the effective electro optic coefficient, and E is the applied electric field. The change in the index of the polymer gives rise to a change in the mode index of the composite waveguide (Δneff)
where Γ is the efficiency factor for the structure with Γ=1 if the optical mode were entirely within the EO polymer cladding. To achieve a high Γ, the majority of optical mode would need to reside within the polymer cladding. However, a waveguide structure with a very high Γ, such as Γ>0.8, would not be practical because the optical mode size would be quite large, resulting in high bend loss and would require widely spaced electrodes. We have calculated the optical mode profiles and the resulting Γ for our basic modulator structure with a range of waveguide dimensions. The modes are computed by direct diagonalization of Maxwell’s equations for the transverse electric and magnetic fields. The mode area is computed by normalizing the transverse intensity to unity and integrating the normalized intensity on a rectangular grid with intensity greater than the threshold. The gamma factor is computed by observing the change in the mode effective index as a function of a change in the index of refraction of the cladding.
As seen in Fig. 4(a), there is a universal relationship for Γ for different waveguide dimensions as a function of optical mode area as measured by the area confining 90% of the optical mode. Very high Γ’s (Γ>0.8) are always associated with large optical mode sizes independent of the waveguide dimensions. From Fig. 4(b) it is clear that small waveguide cross sections lead to higher Γ’s due to poorer optical confinement. A Γ of ~0.7 can be accomplished with silicon nitride waveguide with a properly designed geometry by optimizing the waveguide cross sectional area (<0.23 µm2) and optical mode volume (<2 µm2 90 % mode volume) such that the mode is moderately confined.
A micro-ring resonator based on the composite silicon nitride core with silicon oxide and EO polymer cladding was fabricated with a radius of 28 microns, waveguide width of 0.9 µm and a gap of 0.3 µm between the bus waveguide and ring. Cu electrodes are spaced with a 3 micron gap surrounding the ring and filled in with the EO polymer and poled using the Cu electrodes.
The capacitance of the resonator was measured <10 fF for a ring of radius 28 µm and an electrode gap of 3 µm. The properties of the passive ring resonator were measured using the TE mode of an external cavity tunable laser and without any applied field to the control electrodes (see Fig. 5). The Q was 5,500, the extinction depth 12 dB and the FSR 4.7 nm.
To observe the change in the effective index with applied field, a DC voltage was applied to the poled ring resonator modulator and a wavelength scan made to measure the optical transmission. The resonance was observed to shift 3 pm/V (Fig. 6). Using a calculated value of Γ=0.2 corresponding to a waveguide cross sectional area of 0.41 µm2 and taking into account that the electrodes only surround 80% of the ring, this gives us an effective electro optic coefficient (reff) of ~30 pm/V. We have extracted a range reff from 25–40 pm/V using this EO polymer film in similar devices.
To compare the performance of the EO polymer in our device to conventional methodologies, The EO coefficients (r 33 values) were measured using the Teng-Man simple reflection technique at the wavelength of 1310 nm. For sample preparation, EO films were spin cast onto an indium tin oxide (ITO) glass substrate. [11,12]. Then, a thin layer of gold was sputtered onto the films as a top electrode for contact poling. The poled films of AJTB141/APC exhibited relatively large r 33 value of 58–63 pm/V.
We attribute the lower effective EO coefficient found in our ring resonator modulator compared to the samples used for the Teng-Man measurement to un-optimized poling conditions in the ring resonator device and or a difference in the geometry between the sandwich structures used for the Teng-Man measurement and the co-planar electrodes in the ring resonator modulator.
High frequency measurements were made on the modulator described above by applying an AC field and using an AC coupled 12 GHz New Focus photoreceiver to detect the optical modulation. The laser wavelength was scanned until a maximum in modulation depth was observed and then fixing the wavelength at that point. Using a 2.7 Vpp clock signal with 50 % duty cycle, modulation at 10 GHz is seen in Fig. 7. Modulation depth is approximately 1 dB at this operating voltage. Higher modulation depths are observed with higher drive voltage.
There are several areas where the modulator performance can be improved. The calculated Γ of the device was 0.2. By shrinking the waveguide cross sectional area to 0.23 µm2 (0.45 µm thick×0.5 µm wide), we were able to make ring resonator modulators with higher calculated Γ~0.65. We have measured a shift of the resonance peak as high as 7 pm/V for the same EO polymer material with these higher Γ compared to 3 pm/V for the Γ=0.2 devices. Unfortunately, the Q and extinction depth were poor on this ring resonator, such that the modulation depth was weaker than the device with the lower Γ. We expect to be able to optimize the Q and extinction depth of the higher Γ modulators by depositing a thinner silicon nitride waveguide layer to improve edge roughness induced during etching and by optimizing the gap between the bus waveguide and ring.
Another improvement could be realized by increasing the electro-optic coefficient. As mentioned above, the calculated reff was ~1/2 of the expected value. We believe that optimization of the poling process can reduce that mismatch. In addition, incorporating higher r33 materials into the device should provide improved device performance. The material used in this study has a modest r33 and was chosen for its processiblity and because it has been previously well characterized. We plan to incorporate higher r33 materials in the future to significantly improve the modulator performance.
Finally, reducing the electrode spacing could further improve the device performance. This would only be possible by a change in the waveguide core materials to a higher index material. To meet the back end temperature requirement, amorphous Si would be a possible candidate. By incorporating these improvements, high modulation depth modulators with 1 V drive should be achievable for high data rates and low power.
Electrical chip to chip interconnects are becoming increasingly more difficult to make with high enough bandwidth and a low power consumption. Optical interconnects offer a potential solution, as long as consideration is made to make them with power efficiency in mind. Our “photo-CMOS” approach places monolithic optical components directly on top of the existing CMOS circuitry. Towards this goal, we have fabricated compact micro-ring resonator modulators based on a composite structure consisting of a silicon nitride waveguide on silicon oxide and an EO polymer top cladding. Damascene Cu electrodes are used to electrically control the optical modulations. We have demonstrated a high frequency modulator with modulation observed at 10 GHz and low drive voltage, 2.7 Vpp. Improvement paths to operate at >40 Gb/s at low voltage and low power are outlined.
References and links
1. M. J. Kobrinsky, B. A. Block, J. F. Zheng, B. C. Barnett, E. Mohammed, M. Reshotko, F. Roberston, S. List, I. Young, and K. Cadien, “On-Chip Optical Interconnects,” Intel Technol. J. 8, 129–141 (2004).
2. B. E. Nelson, G. A. Keeler, D. Agarwal, N. C. Helman, and D. A. B. Miller, “Wavelength division multiplexed optical interconnect using short pulses,” IEEE J. Sel. Top. Quantum Electron. 9, 486–491 (2003). [CrossRef]
3. K. Y. Jen, J. Luo, T. D. Kim, B. Chen, S. -H. Jang, J. -W. Kang, N. M. Tucker, S. Hau, T. Yanqing, J. -W. Ka, M. Haller, Y. Liao, B. Robinson, L. Dalton, and W. Herman, “Exceptional electro-optic properties through molecular design and controlled self-assembly,” Proc SPIE 5935, 49–61 (2005).
4. D. Chen, H. R. Fetterman, A. Chen, W. H. Steier, L. R. Dalton, W. Wang, and Y. Shi, “Demonstration of 110 GHz electro-optic polymer modulation,” Appl. Phys. Lett. 70, 3335–3337 (1997). [CrossRef]
5. Y. Shi, W. Wang, J. H. Bechtel, A. Chen, S. Garner, S. Kalluri, W. H. Steier, D. Chen, H. R. Fetterman, L. R. Dalton, and L. Yu, “Fabrication and characterization of high-speed polyurethane-disperse red 19 integrated electrooptic modulators for analog system applications,” IEEE J. Sel. Top. Quantum Electron. 2, 289–299 (1996). [CrossRef]
6. M. R. Reshotko, B. A. Block, B. Jin, and P. Chang, “Waveguide Coupled Ge-on-Oxide Photodetectors for Integrated Optical Links,” Proceedings of IEEE/LEOS 5th International Conference on Group IV Photonics (IEEE, 2008) pp. 182–184.
7. M. -C. Oh, H. Zhang, C. Zhang, H. Erlig, Y. Chang, B. Tsap, D. Chang, A. Szep, W. H. Steier, H. R. Fetterman, and L. R. Dalton, “Recent advances in electrooptic polymer modulators incorporating highly nonlinear chromophore,” IEEE J. Sel. Top. Quantum Electron. 7, 826–835 (2001). [CrossRef]
8. B. M. A. Rahman, V. Haxha, S. Haxha, and K. T. V. Grattan, “Design optimization of polymer electrooptic modulators” J. Lightwave Technol. 24, 3506–3513 (2006). [CrossRef]
9. B. -J. Seo, S. Kim, H. Fetterman, W. Steier, D. Jin, R. D, and Dinu, “Design of Ring Resonators using Electro-Optic Polymer Waveguides,” J. Phys. Chem. C 112, 7953–7958 (2008) [CrossRef]
10. P. Rabiei, W. H. Steier, C. Zhang, and L. R. Dalton, “Integrated WDM polymer modulator” Optical Fiber Communications Conference. (OFC). Postconference Technical Digest, (IEEE, 2002) pp. 17–22.
11. C. C. Teng and H. T. Man, “Simple reflection technique for mearsuring the electro-optic coefficient of poled polymers,” Appl. Phys. Lett. 56, 1734–1736 (1990). [CrossRef]
12. D. H. Park, C. H. Lee, and W. N. Herman, “Analysis of multiple reflection effects in reflective measurements of electro-optic coefficients of poled polymers in multilayer structures,” Opt. Express 14, 8866–8884 (2006). [CrossRef] [PubMed]