1. Introduction

Low-power optical links between complementary metal oxide semiconductor (CMOS) chips have been studied extensively, and ultrasmall lasers, modulators, and photoreceivers with good integrability have been reported [1–5]. In moving beyond simple optical links, the photonic network on chip (PhNoC), which includes many integrated nanophotonic devices that can manage high-speed optical processing, is also needed for more intelligent photonic/CMOS collaborative data processing. For these kinds of systems, low-power electrical-to-optical and optical-to-electrical conversion is required, because they might be a bottleneck in optical links in terms of power consumption and communication bandwidth. It has been discussed that future interconnect technology will demand an energy cost of less than 10 fJ/bit for a single optical link [6]. Photodetectors (PDs) will be key components to meet the demand, because optical sensitivity determines the power consumption of transmitters (laser diodes and modulators) and the overall optical link required for signal communication.

Some photoreceivers based on Si waveguides containing a Ge absorber have already been reported for the purpose of CMOS integration [2, 3]. The on-chip integration of a modulator based on a silicon ring resonator and a Ge-based PD has also been reported at an energy cost of 120 fJ/bit (not including the power of the laser source) for 3 Gb/s data transmission [4]. We have also demonstrated an optical link between a PhC-based nanolaser and a detector on InP with an energy cost of 17.3 fJ/bit for 5 Gb/s data transmission [7]. These results effectively suggest the feasibility of an inter/intra-chip optical link. However, these energy costs were mainly evaluated for the transmitter side and did not include the receiver side. Note that actual optical links generally include a trans-impedance amplifier (TIA) and a limiting amplifier (LA) that are connected to a PD, and these electrical amplifiers dominate the power consumption at a several tens of mW level [3]. This would amount to a pJ/bit level energy cost if we assume a bit rate of 10 Gb/s, and this situation is therefore raising concerns as it constitutes a significant bottleneck in optical links.

If we are to solve this problem, one of the challenges with PDs that we must overcome is to realize a small capacitance. If the capacitance can be greatly reduced to an fF level or less by reducing the size of the PDs, the RC constant could be kept at a low level even during connection to a high load resistor. The result would be a low thermal noise and an enhanced output voltage while maintaining a fast response. This would lead to the reduction of electrical amplification or even its elimination (referred to as a receiver-less PD [6]), which would greatly reduce the overall power consumption of an optical link. There would then be a strong demand for nano-PDs with a small junction capacitance and high absorption efficiency. A simple estimation indicates that a nano-PD with a 1 fF capacitance would be capable of integration with a 10-kΩ load resistor without a TIA and thereby provide an output voltage of 0.2 V for an optical input with sub-fJ energy and over 10 Gb/s repetition.

Plasmonic structures are candidates for nano-PDs, because they are extremely small and close to the CMOS scale [8, 9]. However, they intrinsically have a significant excess loss due to the light absorption by metal, and efficient light coupling into a nano-PD and high quantum efficiency are still challenging problems with regards to their applicability for realistic receivers. Semiconductor PhC waveguides and nanocavities show greater promise as nano-PDs because of their strong optical confinement in ultrasmall dimensions, which can be well realized by the current fabrication process. We have already reported a silicon-based PD using a PhC nanocavity, which enabled the detection of 1.55 μm wavelength 1ight by two-photon absorption (TPA) in a nanocavity [10]. However, it requires an ultrahigh Q factor to enhance the TPA efficiency, and therefore the operation bandwidth was strictly limited. Recently, we developed an ultracompact buried heterostructure (BH) consisting of InGaAsP in a thin InP-PhC waveguide by using the etching-and-regrowth technique [11]. We have already demonstrated a current-injection PhC nanolaser that achieved a record low threshold under room temperature CW conditions [5, 12]. Such a BH technique should provide good applicability for novel InGaAs-based nano-PDs, because unlike any other PD, this structure can confine both photons and carriers in an ultrasmall space. Hence it might enable us to fabricate InGaAs nano-PDs with ultrasmall capacitance by reducing the size of p-i-n junctions. In addition, compared with Ge-based PDs, InGaAs has a higher absorption coefficient and a longer applicable wavelength region such as the L-band range, which should allow us to realize a wide-range wavelength-division multiplexing (WDM) system. It can also make it possible to integrate with InP-based active nano-photonic elements such as all-optical switches and memories [13, 14], which would be attractive for realizing a future PhNoC.

In this paper, we report the first nano-PD based on a PhC waveguide including an ultracompact InGaAs absorption layer with a length of 3.4 μm, and demonstrate a high DC responsivity of ~1 A/W and a dynamic response with a 10-Gb/s eye pattern. While our PhC-based PD has high efficiency and a fast response, it is also much smaller than previous InGaAs-based waveguide detectors [15, 16] because of the strong confinement of the photons and carriers. This can also be effective in achieving a huge reduction of the junction capacitance to less than 1 fF, which suggests the potential for realizing photoreceivers that do not need electrical amplifiers. Such ultrasmall nano-PDs should offer low-power optical links and optical processing in combination with CMOS electronics.

2. Device design and fabrication

Figure 1(a) is a schematic of our nano-PD based on a PhC waveguide, which consists of an InGaAs absorber embedded in an InP-PhC line-defect waveguide and a lateral p-i-n junction. A PhC can confine light in a small core waveguide, which means that high responsivity can still be expected for a PD only a few μm long. This structure also allows us to achieve strong carrier confinement in the InGaAs absorber because of the bandgap offset with the surrounding InP. Therefore, it enables us to reduce the length of the p-i-n junction and subsequently reduce the junction capacitance.

 

Fig. 1 Structure and SEM of the PD. (a) Structural schematic of PhC-based nano-PD. (b) SEM image of fabricated device. This is a different sample from the measured one, and the length of the embedded InGaAs absorber is shorter in the image.

Download Full Size | PDF

Our PD was fabricated using the same procedure that we reported in [17]. An InGaAs bulk absorber and an InGaAs sacrificial layer were grown on a semi-insulated InP substrate. The composition of both layers is In_0.53Ga_0.47As for lattice matching with InP. Butt-joint regrowth was performed and the InGaAs absorber was embedded in a 250-nm-thick InP slab. To fabricate a lateral p-i-n junction, we employed Zn diffusion and Si ion implantation, respectively, for the p- and n-type doping of an undoped InP layer. The doping concentration for these doped layers is approximately 1 × 10¹⁸ cm⁻³. PhC airholes were formed by EB lithography and dry etching. Metal was deposited for electrical contacts. Finally, the InGaAs sacrificial layer was etched to form an air-bridge structure. The separation between the p- and n-doped layers was designed to be 1.5 μm, but it will decrease during the process. This separation determines the carrier transit time across the depletion region, and has not yet been optimized to enhance the operation bandwidth. Figure 1(b) shows a SEM image of the fabricated sample, indicating a flat surface resulting from the successful butt-joint regrowth process. The airhole diameter and the lattice constant of the PhC were 225 and 420 nm, respectively. Because of the index difference between the InP waveguide and the InGaAs-embedded waveguide, their widths should be adjusted to match their waveguiding bands. To this end, the widths of the InP and InGaAs-embedded region were changed to 1.1W₀ and 0.95W₀, respectively, where W₀ is the basic line defect width defined as the removal of one row of air holes.

Since we can expect an absorption coefficient of around 10000 cm⁻¹ for a wavelength of 1.55 μm [18] and a confinement factor (which is defined as the ratio of the light intensity of the propagation mode within the InGaAs absorber to the total intensity over the entire space) exceeding 0.5, a length of 3 μm and a single round trip are sufficient for an absorption efficiency of more than 90% (excluding the coupling loss between the InP and InGaAs-embedded waveguides). The InGaAs absorber has a width of roughly 0.3 μm and a height of 0.15 μm in its cross-section, and a length of 3.4 μm to ensure complete absorption of the light. The p-i-n junction was designed to be 4.2 μm long. Hence the junction capacitance was estimated to be 0.12 – 0.23 fF by assuming the p-i-n junction to be a parallel-plate capacitor (approximately calculated from C = ε₀ε_InPS/d, where ε₀ is the permittivity of a vacuum, ε_InP = 12.4 is the relative permittivity of InP, S = 4.2 μm × 0.25 μm is the cross section of the doped region, d = 0.5 – 1.0 μm is the roughly estimated gap between the p and n regions).This is an order of magnitude lower than previously reported Ge-based p-i-n detectors [2], and may allow a high load resistance of 10 kΩ for the receiver with an RC time constant of less than 15 ps.

3. DC responses

First, the photocurrent characteristics for a continuous-wave (CW) light input were measured to evaluate the DC responses. For the measurement, a fiber polarization controller was used to tune the input light to TE polarization. Figure 2(a) and 2(b) show the transmission spectra for the input waveguides and the photocurrent spectrum of the PD, respectively, for which the light wavelength was swept and the generated carriers were collected by applying reverse bias voltage. As shown in Fig. 2(b), we confirmed that an efficient photocurrent was realized for a 1.46 – 1.49 μm wavelength range. The waveguiding bands of the input InP waveguide and InGaAs-embedded waveguide are located below the wavelengths of 1.52 and 1.49 μm, respectively. In fact, we expected a wavelength of around 1.55 μm to be available for the device, but the airhole diameter became larger than we anticipated, thus shifting the operative waveguide band down to 1.49 μm. The disappearance of the photocurrent below 1.46 μm is due to the cut-off of the W8 waveguide, as shown in the transmission spectrum in Fig. 2(a). The operation wavelength range can be easily shifted by appropriately determining the airhole size, enabling us the capability for C + L band wavelength range (1520 – 1620 nm) or beyond. The photocurrent spectrum exhibits periodic peaks with a 6.1 nm interval, which appears as a result of the Fabry-Perot (FP) resonance between the facet end of the waveguide and the input boundary of the PD. As discussed later, the PD does not itself have a resonant effect for enhancing the absorption efficiency, but the FP resonant filter in front of the PD compensates for the degradation in absorption efficiency caused by the optical reflection at the input boundary of the PD. In future designs, the optical coupling structure at the boundary should be optimized to suppress the FP resonance and obtain a flat photocurrent spectrum. In addition, a small photocurrent was observed even above the waveguide band. This resulted from the light absorption at the InGaAs sacrificial layer, which could be detected because the light was scattered at the input waveguide facet and the generated carriers were swept through the sacrificial layer and the InP substrate.

 

Fig. 2 Photocurrent spectrum for CW light input. (a) Transmission spectra for W8 waveguide and W8 + W1 waveguide. (b) Photocurrent spectrum. Input CW power was −17.3 ± 1.3 dBm. The inset is a schematic showing that the generated carriers at the InGaAs sacrificial layer are swept through the InP buffer layer.

Download Full Size | PDF

The photocurrent for a different reverse bias voltage and CW optical power is shown in Fig. 3(a), for which the light wavelength was set at the peak of the photocurrent spectrum. The dark currents were approximately 2 and 40 nA when the reverse bias voltages were set at −1 and −4 V, respectively. This originated from the leakage current that passed through the sacrificial layer and the substrate. This will fall to much less than 1 nA if an InAlAs is used for the sacrificial layer to block the leakage current due to the large bandgap energy [7]. To estimate the on-chip DC responsivity, we varied the input light power from −63.5 to −13.5 dBm in 5 dB steps, for which we estimated that the optical power injected into the InP-PhC waveguide was −11.1 dB by measuring the power in the optical fiber and the coupling loss into the InP-PhC waveguide. The on-chip DC responsivity was then estimated to be 0.56 – 1.00 A/W at a bias voltage of −1 V. A responsivity of 0.56 A/W was estimated at the bottom wavelength of the FP resonance (1470 nm), for which the optical transmission into the PD was strictly limited by the reflection at the input edge of the PD. On the other hand, a responsivity of 1.00 A/W was estimated at the peak wavelength of the FP resonance (1473 nm), for which the optical transmission into the PD becomes high because of the FP resonant effect. Note that our PD is different from the well-known resonant-cavity-enhanced (RCE) PDs [19], which are formed by making the resonator the absorber region. The light launched into the PD region should be mostly absorbed (> 90%) with a single trip in the PD, so our PD does not itself have an RCE effect. In fact there is optical reflection at the input boundary of the PD, but the absorption efficiency degradation caused by the reflection is compensated for by the FP-resonator filter in front of the PD. Importantly, these results suggest that the light absorption and carrier collection were efficient enough for a 3.4-μm-long InGaAs absorber. Note that the enhancement of the non-radiative recombination due to the carrier trapping at the hetero interface was not apparent, which is unique and significant for our BH structure. Since the optical reflection at the input boundary of the PD is still the main problem limiting responsivity, particularly when we consider the case where there is no FP resonant effect, it should be reduced by further optimization of the optical coupling structure into the PD.

 

Fig. 3 DC responses. (a) Photocurrent versus applied bias voltage characteristics for CW input light. The wavelength was set at the peak of the photocurrent spectrum (1473 nm). The different colors denote the different optical input powers launched into the PD. (b) Photocurrent versus optical input power characteristics plotted for a bias voltage at −1 V. The red and blue plots are the results for the peak wavelength (1473 nm) and bottom wavelength (1470 nm), respectively.

Download Full Size | PDF

4. Dynamic response

We investigated the dynamic performance of our PD by injecting an intensity-modulated optical signal under a reverse bias through a bias tee thereby extracting an RF signal. Figure 4 shows the operation dynamics at the photocurrent peak wavelength (1473 nm) in Fig. 2(b). We obtained an eye opening for a 10-Gb/s non-return-to-zero (NRZ) signal generated with a 2³¹−1 pseudo-random bit sequence, as shown in Fig. 4(a). Since the TIA was not used for observation with the sampling oscilloscope, the eye pattern was slightly noisy and the bit error rate could not be evaluated. We also confirmed that the light outside the waveguiding band (1550 nm) did not generate any photoresponse. Figure 4(b) shows the small signal responses for different reverse bias voltages. The 3-dB bandwidth was 6 GHz when the bias voltage was −4 V. Several factors may be involved in limiting the operation bandwidth, and these might include a long carrier transit time across the depletion region and an RC time limitation caused by a parasitic capacitance except for the small p-i-n junction. The former and latter problems will be effectively solved by optimizing the separation between the p- and n-doped regions, and using an appropriate wafer composition to suppress the excess charging, respectively. A higher speed response can be expected by dealing with these issues, and a further reduction of the capacitance and subsequent improvement of the RC time constant might also be expected by employing an ultrasmall PhC nanocavity and decreasing the size of the p-i-n junction.

 

Fig. 4 Dynamic responses. (a) Eye pattern for 10 Gb/s NRZ optical signal. Reverse-bias voltage was 7 V. (b) Small signal responses for different reverse-bias voltages. The wavelength was set at the peak of the photocurrent spectrum.

Download Full Size | PDF

5. Summary

We successfully fabricated a 3.4-μm-long InGaAs PhC nano-PD, and confirmed a DC responsivity of ~1 A/W, an eye pattern for a 10 Gb/s signal, and a 3 dB bandwidth of 6 GHz. This is the first demonstration of such a small, efficient InGaAs nano-PD based on a PhC waveguide, and it has potential for use in integrable low-power photoreceivers because of its small size for a small junction capacitance. Nano-PDs based on nanocavities can also be expected that will offer further size reduction while maintaining high responsivity, thus enabling extremely small junctions to be connected with high load resistors. Our nano-PD can be easily connected with PhC nanolaser sources that have a similar BH structure on the same substrate and can operate in an fJ energy regime. It can realize a low-power photonic layer and is therefore a good candidate for constructing a green optical link for CMOS chips and PhNoC.