1. Introduction

Optical interconnect technology is expected as a key solution to solving major performance limitations in high performance computing caused by bus bandwidth bottleneck, latency and power consumption issues [1–5]. A critical component in an optical interconnect is a high speed silicon optical modulator with low power/energy consumption and a small footprint [6]. While Mach-Zehnder interferometer (MZI) silicon modulators have been developed with speed up to 30 GHz, they usually exhibit a large footprint, have energy consumption greater than 5 pJ per bit, and require a large driving voltage of 6-8 V [7–9]. These make them not suitable for the applications of chip-level optical interconnect. Promising compact and low power modulator candidates include carrier-injection silicon microring modulators [10, 11], carrier-depletion microdisk and microring modulators [12, 13], electro-absorption GeSi modulators using the Franz-Keldysh effect [14, 15], and GeSi quantum well modulators [16]. The spectral bandwidth of microcavities is usually less than 1 nm and that of GeSi modulators is about 15 nm. Therefore, all these modulators suffer from a narrow bandwidth, which limits possible data bandwidth enabled by wavelength multiplexing techniques. The need for wavelength-tunable microcavity modulators is even greater due to the uncertainty in resonance wavelength arising from tolerances in the fabrication process [6]. Here we demonstrate a high speed silicon microring modulator with a local microheater. The local heater can heat up the ring and red-shift the resonance, with a tuning efficiency of 2.4 mW/nm (2.4 mW heating power is needed to shift the resonant wavelength by 1 nm). High speed modulation is achieved by carrier depletion in a reverse-biased pn junction embedded in the ring; the modulator exhibits a drive voltage of 3 V, a high speed of 12.5 Gbps, ultralow modulation energy consumption of 40 fJ per bit (not including tuning power), and a small ring radius of 5 µm.

2. Design and fabrication

The use of microrings or microdisks can dramatically reduce modulation power and device size by exploiting optical field confinement in a small area. Further advantages include high modulation ratios, low insertion losses, and low V_pp (peak-to-peak swing voltage) [13]. The microring modulator presented here consists of a ring resonator coupled to a neighboring bus waveguide. The bus and ring waveguides have a width of 0.5 µm, a height of 0.25 µm and a slab height of 50 nm [Fig. 1(a) ]. The ring radius is 5 µm and the ring-bus gap is 150 nm. Tuning the effective index of the ring waveguide modifies the resonant wavelength which induces a strong modulation of the transmitted signal at the resonance wavelength. Reverse-biased pn diodes or forward-biased p-i-n diodes have been demonstrated to realize high speed modulation of silicon waveguides [8–13] through plasma dispersion effects [17]. Here, we designed an asymmetric lateral pn junction with a p doping concentration of 5x10¹⁷ cm⁻³ and an n doping concentration of 1x10¹⁸ cm⁻³ [13]. We fabricated the device on a silicon-on-insulator (SOI) platform with a 3-µm-thick buried oxide. Following waveguide trench etching, multiple lithography steps were used to define various doping areas. High-energy ion implantation and rapid thermal annealing defined the pn diode and the heavy doping regions. A 1.2 µm-thick oxide was deposited as the top cladding layer. A 100 nm-thick Ti film was then deposited on top of the cladding and along the ring, which acts as a local heater. The heater has a width of 1 µm. Figure 1(b) shows a top-view scanning electron microscopy (SEM) image of a fully fabricated ring modulator.

 

Fig. 1 (a) Cross-section of the designed modulator with a local heater. (b) Top-view SEM image of the fully fabricated microring modulator. The ring radius is 5 µm.

Download Full Size | PDF

3. Measurement

We characterize the heater efficiency by measuring the transmission spectra under different heating powers, as shown in Fig. 2(a) . As we increase the electrical power of the metal heater, the temperature of the ring waveguide rises, which produces a resonance red-shift (silicon has a thermal coefficient of ~1.86x10⁻⁴/°C). The resonant wavelength is extracted as a function of heating power, presented in Fig. 2(b). The linear fitting shows that the power to shift the resonance of 1 nm is 2.4 mW. Since the free spectral range (FSR) of this ring is ~19.5 nm, the power to tune one whole FSR is about 46 mW. This tuning power is comparable to that reported for passive silicon ring resonators [18] but higher than for recently reported optimized heaters [19, 20]. Compared with passive only rings where the silicon substrates are the main heat sink, the metal contacts of the pn junction also dissipate heat, which reduces the heating efficiency. From Fig. 2, it is seen that the extinction ratio (ER) of the ring resonance varies with ring temperature, however, a >10 dB ER is achievable over the heating range. This ER variation may be due to the coupling variation between the bus waveguide and the ring or the ring waveguide loss variation arising from the opto-elastic effect.

 

Fig. 2 (a) Transmission spectra of the ring resonator with different heating power. (b) Resonant wavelength versus heating power and its linear fitting.

Download Full Size | PDF

The high speed modulation is achieved by modulating the reverse bias of the pn junction. We measured the transmission spectra under different bias voltages, as shown in Fig. 3 . The quality factor of the ring resonator is estimated at ~9000. From the value of the quality factor, we estimate that the ring waveguide has a propagation loss of ~45 dB/cm, which comes from the doping and also from the sidewall roughness. The resonance shift per unit voltage is 16.5 pm/V. A V_pp of 3V is sufficient to achieve a modulation depth of >6 dB. Considering an experimentally measured junction capacitance of ~17 fF for our device and a V_pp of 3 V, we can obtain the energy consumption as ~40 fJ/bit [13].

 

Fig. 3 Transmission spectra of the ring at different bias voltages.

Download Full Size | PDF

In order to verify the feasibility of high temperature operation, we carried out eye diagram measurement with different heating powers. The modulator was driven by an NRZ PRBS 2²³-1 signal with a V_pp of 1.5 V and a d.c. bias of −1 V through a high speed probe. Due to microwave reflection from the modulator to the probe, the actual V_pp applied on the ring modulator is expected to be 3 V. The optical output was amplified by an erbium doped fiber amplifier (EDFA) and then measured using a high-speed oscilloscope equipped with an optical detector. The measured eye diagrams exhibit clear eye openings for 12.5 Gbps with a modulation depth of about 6 dB without any heating (i.e., at room temperature), as shown in Fig. 4(a) . The noise in the eye diagrams may come from the EDFA and/or the microwave reflection. We then applied a heating power of 18 mW, which tunes the ring resonance from 1550.7 nm to 1558.0 nm. An eye diagram with an input wavelength of 1558.0 nm was then collected, shown in Fig. 4(b). Since the resonance shift can be expressed by$\mathrm{\Delta }\lambda =\frac{\lambda }{n_g}\left(\frac{\partial N_{eff}}{\partial T}\right)\mathrm{\Delta }T$, where N_eff is the effective index and n_g is the group index, the theoretical resonance shift efficiency is $\frac{\mathrm{\Delta }\lambda }{\mathrm{\Delta }T}\approx 0.078\text{\hspace{0.17em} nm/}°\text{C}$. The temperature increase of the ring with 18 mW heating power is therefore expected to be ~93 °C for the ring waveguide. Comparing Fig. 4(b) with Fig. 4(a), we do not see detrimental effects due to this high temperature operation. The current density in the heater is estimated about 3.5 MA/cm², for which electro-migration effects may occur to degrade the heater performance. Careful design of heater width and thickness and/or the use of other heater metals/alloys may be needed to avoid this effect. In addition, the resistivity increase of our Ti heater with respect to temperature was measured as ~0.4% per degree.

 

Fig. 4 (a) 12.5 Gbps modulation of the device without any heating power at a wavelength of 1550.70 nm. (b) 12.5 Gbps modulation of the device with 18 mW heating power at a wavelength of 1558.00 nm. The temperature increase is expected to be ~93 °C.

Download Full Size | PDF

4. Discussion and conclusion

The thermal tuning power would not increase if the modulation speed increases. From the quality factor (~9000) and series resistance of our device (measured as ~100 ohm), our ring modulator should be able to operate at 20 Gbps [13]. Recently reported ring modulator by other groups also demonstrates a high speed 3 dB bandwidth of 19 GHz [21]. Therefore, if adding a thermal tuning power of 12 mW which can shift the resonance by ~5 nm (5 nm would allow a temperature tolerance of ~65 °C), the total energy consumption would increase to ~650 fJ/bit under 20 Gbps operation. More efficient heating configurations may be designed using techniques such as thermal isolation beneath modulators [20] or using waveguides as a heating resistor [19] (~0.55 mW/nm efficiency has been demonstrated). This would bring down the total energy consumption of modulation and tuning to under 500 fJ/bit. Further power reduction can be realized by reducing the ring radius since the tuning efficiency is usually proportional to the inverse of the ring radius (microrings with a 1.5 µm radius and with similar waveguide geometry have been demonstrated in Ref [22].). Even including this tuning power, the energy consumption of ring modulators is still significantly lower than that of MZI modulators which is usually larger than 5 pJ/bit. In addition, as demonstrated in Ref [20], 15 µm spacing is sufficient to avoid thermal crosstalk if more ring modulators are needed on a single chip, resulting in much more compact devices than MZI modulators.

In conclusion, a compact (5 µm radius ring), high-speed (12.5 Gbps) silicon electro-optic microring modulator with a low V_pp (3V) has been demonstrated with tunable resonance accomplished by means of a local microheater. This modulator consists of a ring resonator as an optical platform and realizes index modulation by using a reversed-biased lateral pn diode embedded in the ring. The heater on top of the ring can tune the resonance wavelength with an efficiency of 2.4 mW/nm. This device aims to solve the narrow bandwidth problem of silicon microcavity modulators and increase the data bandwidth in optical interconnect systems. To the best of our knowledge, this is the first experimental realization of microring modulators integrated with local metal heaters, although passive silicon rings with heaters have been demonstrated [18–20]. The metal heaters enable wide tuning ranges, much larger than that of forward biased p-j-n junction based heaters demonstrated in Ref [23]. In addition, we would like to stress that athermal operation of ring resonators and couplers have been demonstrated [24,25], however, submicron silicon waveguides are very sensitive to fabrication variations and athermal operation limits the application of the thermo-optic effect to compensate the wavelength variation arising from fabrication tolerance [20].