Extrinsic Photodiodes for Integrated Mid-Infrared Silicon Photonics

Silicon photonics has recently been proposed for a diverse set of applications at mid-infrared wavelengths, the implementation of which require on-chip photodetectors. In planar geometries, dopant-based extrinsic photoconductors have long been used for mid-infrared detection with Si and Ge acting as host materials. Leveraging the dopant-induced sub-bandgap trap-states used in bulk photoconductors for waveguide integrated mid-infrared detectors offers simple processing, integration, and operation throughout the mid-infrared by appropriate choice of dopant. In particular, Si doped with Zn forms two trap levels ~ 0.3 eV and ~ 0.58 eV above the valence band, and has been utilized extensively for cryogenically cooled bulk extrinsic photoconductors. In this letter, we present room temperature operation of Zn+ implanted Si waveguide photodiodes from 2200 nm to 2400 nm, with measured responsivities of up to 87 mA/W and low dark currents of<10 microamps.

heterogeneous integration schemes present an inherent difficulty by imposing constraints on material quality and process integration. Extrinsic detectors, which utilize absorption transitions from dopant-induced trap states within the bandgap of a host material, present a simple solution for high-performance integrated mid-infrared PDs and alleviate the need for heterogeneous integration. These PDs can potentially be integrated into a standard CMOS process flow by adding an ion implantation and annealing step after activation of the source and drain implants, and prior to the deposition of back-end dielectrics and interconnect metallization.
Alternatively, this additional fabrication step can be performed as a post-process, as is done here (see Methods). The Si dopants used in bulk photoconductors can potentially be integrated for detection from a range of wavelengths from 1.5 μm to greater than 25 µm (Fig. 1a), while doped Ge bulk photoconductors have been demonstrated at wavelengths greater than 100 µm 3 .
For the 2.2 µm to 2.4 µm wavelength range Zn, Se, and Au dopants have been shown to produce a suitable defect level in Si 5 , and planar photoconductive detectors based upon these dopants have been demonstrated 3,7 . While these detectors required liquid nitrogen cooling, it has very recently been shown that silicon hyperdoped with Au can generate photocurrent up to a wavelength of 2.2 µm at room temperature with a Schottky contact configuration 17 . Since cooling is not required to eliminate dark current in diode-based detectors, room temperature operation is achieved with the PDs explored here as well.
Although transition metals such as Au and Zn are generally not CMOS compatible, as they can diffuse into Si and adversely affect carrier mobility, the other dopants in Fig. 1a do not have these restrictions. For the wavelength range explored here, Si:Se could be substituted for CMOS compatibility, and has been demonstrated for a SiWG PD operating at 1.55 μm 18 .
The PDs demonstrated here are based on a p-i-n diode structure fabricated in a 250 µm long Si rib/ridge waveguide with 520 nm × 220 nm channel section and a 50 nm doped silicon ridge used to form ohmic contacts to Al (see Methods), as shown in Figs. 1b,c. The intrinsic region of the p-i-n diode corresponds to the channel section of the SiWG, and is implanted with Zn + to form the final p-Si:Zn-n PD structure.
Two Zn + implantation doses are investigated, 10 12 cm -2 and 10 13 cm -2 , corresponding to estimated average Zn concentrations inside the channel section of the waveguide of NZn = 4.5×10 16 cm -3 and NZn = 4.5×10 17 cm -3 , respectively. The Zn concentrations are estimated by averaging the stopping range of Zn + inside of the Si device layer, which is calculated by SRIM software 19 for an acceleration voltage of 260 keV. The acceleration voltage is chosen for maximum overlap between the generated trap states and the quasi-TE waveguide mode (see SI). Subsequent to implantation, the PDs were annealed in atmosphere for a series of increasing temperatures, reaching a maximum of 350°C, and the responsivity was found to increase with annealing temperature.
The detection mechanism of our p-Si:Zn-n PDs is due to substitutional Zn atoms in the Si lattice, which act as a double-acceptors and result in the two defect levels ( Fig. 2a) with energy levels of Ed1 ≈ Ev + 0.3 eV and Ed2 ≈ Ev + 0.58 eV 5,6 . While the position of the Fermi level in the Si:Zn region is not known, the excess carrier concentration is below that of the p and n regions, ensuring that the Si:Zn region will be fully depleted with the application of a reverse bias voltage. Photon-induced transitions occurs between Ed2 and the conduction band, which corresponds to a transition energy of Eg -Ed2 = 1.12eV -0.58eV = 0.54eV with a peak absorption wavelength of ≈ 2.3 µm. Photocurrent generation due to this transition is shown to be a single-photon process by the linearity measurements in Fig. 2b. The presence of Ed1 does not contribute to photocurrent generation in the wavelength range of interest; however, its presence should impact the thermally assisted re-population rate of Ed2 and thus the internal quantum efficiency, ηi, of the PD. The responsivity, defined as R = Iph/Pin, where Iph is the photocurrent and Pin is the on-chip optical power entering the PD, is measured as a function of reverse bias voltage with 2.2 µm-wavelength excitation ( Fig. 2c). While the dark current (Fig. 2c, inset) increases with dopant concentration, it remains below taken. The kink in the responsivity curve of the 10 12 cm -2 implantation dose PDs between 12V and 15V is indicative of avalanche multiplication, which has previously been observed in similar SiWG p-i-n contact configurations 20 . The responsivity is largest at shorter wavelengths (Fig. 2d), reaching a maximum of 8.3 ± 2.6 mA/W and 34.1 ± 10.6 mA/W for doses of 10 12 cm -2 and 10 13 cm -2 , respectively, at a wavelength of 2.2 µm and a reverse bias of 20V. For 3 mm-length PDs maximum responsivities of 87 ± 29 mA/W were measured under similar conditions, and the dark current remained below 10μA at 20V reverse bias (see SI). The decrease in responsivity with increasing wavelength is due to parasitic absorption from the contacts (free-carrier absorption in p, n doped regions as well as the Al regions), which increases from > 30% at 2.2 µm to > 89% at 2.4 µm for a dose of 10 12 cm -2 , and from > 10% at 2.2 µm to > 85% at 2.4 µm for a dose of 10 13 cm -2 (see SI). The parasitic absorption has been determined by transmission measurements through an un-implanted diode with the same geometry, p, n doping, and contact metallization (see Methods). By moving the contacts and p, n doped regions further from the waveguide, the responsivity at longer wavelengths can be substantially improved. The modal absorption due to Zn + implantation is determined by transmission measurements with parasitic contact absorption being taken into account (see SI), and is shown in Fig. 3 for both doses. For the PD with a Zn + dose of 10 12 cm -2 , the modal absorption coefficient peaks at a wavelength of 2.325 µm corresponding to a transition energy of 0.533 eV. Similarly, the PD with a Zn + dose of 10 13 cm -2 exhibits a peak in the modal absorption coefficient at a wavelength of 2.250 µm, corresponding to a transition energy of 0.551 eV. These peak absorption wavelengths correspond to the transition from Ed2 to the conduction band shown in Fig. 2a.
Even with parasitic absorption taken into account, the modal absorption coefficient along with the measured responsivities indicate that ηi of the PDs is below 5%, which is significantly lower than the Though certain lattice defects can contribute to photocurrent generation, such as the Si di-vacancy defect which has been used extensively for SiWG PDs 20,24-27 , these particular lattice defects anneal out at temperatures above 300°C 24 . Furthermore, the responsivity of Si di-vacancy PDs decreases rapidly with increasing wavelength 27 , as the position of the trap state has been identified to lie at Edivacancy = Ec -0.4 eV 28 .
Thus, the absorption coefficients measured in Fig. 3 and the post-implantation annealing temperature used indicate contributions to photocurrent by Si di-vacancies is not significant. However, the remaining lattice defects are potentially contributing to optical scattering and increased carrier recombination.
While the number of activated Zn dopants in the PDs is not known, the responsivity was found to increase with increasing annealing temperatures (see Methods), suggesting that activation can be improved further with higher-temperature annealing. Additional implantation parameters can also be optimized, as improved activation and reduced out-diffusion have been achieved by performing Zn + implantation at 350°C 22 . Alternatively, other doping methods can be used to produce higher activation such as pulsed laser melting to supersaturate the Si with Zn 29 . Beyond processing improvements for increased Zn activation, ηi

Methods:
Device fabrication. Un-implanted p-i-n SiWG diodes were fabricated on the CMOS line at MIT Lincoln Laboratory as described in ref. 26 using SOI with a 3 µm BOX and with the device dimensions given in Fig. 1a. Subsequently, contact photolithography was used to define a Shipley S1811 implantation mask at Brookhaven National Laboratory, and Zn + ion implantation was performed in the Ion Beam Laboratory, State University of New York at Albany. An acceleration voltage of 260 keV was used, corresponding to a stopping range of 118 nm into the waveguide as calculated by SRIM software 19 , resulting in the maximum overlap between the induced defect concentration and the fundamental quasi-TE mode of the waveguide (see SI). The ion beam was launched 7° from normal incidence to the sample surface, and implantation current densities of 1 nA/cm 2 and 3-7 nA/cm 2 were used for the 10 12 cm -2 and 10 13 cm -2 implantation doses, respectively. Post implantation, the devices were annealed in atmosphere at 250°C for for 10 minutes per step. Photocurrent measurements were taken between each step, and the device performance was found to improve with each annealing cycle. The annealing temperature was limited to 350°C, since the 100 nm thick Al contacts were found to reflow above this temperature.
Experimental setup. Optical power from an external-cavity Cr 2+ :ZnSe tunable laser (IPG Photonics) is coupled into the SiWG PD using a lensed-tapered fiber (Oz Optics) to a 5µm × 220 nm silicon fan-out taper. Bias voltages are applied and currents are measured using a Keithley 2400 Source/Meter, and the transmitted power through the PD was monitored on a Yokogawa AQ6375 optical spectrum analyzer with the same coupling scheme as the input. The on-chip power entering the PD is determined by measuring the laser output power at each wavelength, and subtracting the loss from propagation through the input lensed-tapered fiber, as well as facet loss from coupling between the lensed-tapered fiber and the silicon fan-out taper. A schematic of the experimental setup is shown in the supporting information.
Determination of error bars. Error bars are determined by uncertainties in losses from the measurement setup (see SI), which include lensed-tapered fiber (LTF) losses and single-mode fiber transmission/bending losses. Additionally, uncertainty in the waveguide facet loss from fan-out tapers couplers have also been taken into account and as well as temporal fluctuations in the laser output as tracked by the transmitted power. Further details are provided in the supporting information.
Extraction of the modal absorption coefficient. Transmission measurements were performed Zn + implanted PDs, and were repeated for an un-implanted diode (having identical p and n implants and contact metallization as the Zn + implanted PD) of the same length. Using these measurements, the effects of parasitic absorption in the Al contacts is taken into account by calculating the modal absorption coefficient, where T is the power transmission in dB and L = 250 µm (Fig. 1c). By subtracting the parasitic absorption coefficient αeff,par, measured from an un-implanted diode, from the absorption coefficient of a Zn + implanted PD, αeff,tot, the absorption coefficient due to defects shown in Fig. 3 is determined: αeff,Zn = αeff,totαeff,par. The contact absorption becomes dominant for λ > 2.325 µm (see SI), resulting in the loss of responsivity with increasing wavelength seen in Fig. 2d.

Measurement setup:
Our measurement setup is shown in Fig. S1. An external cavity Cr 2+ :ZnSe tunable laser (IPG Photonics) passes through an isolator (ISO) and is free-space coupled to a single-mode fiber (SMF) using a fiber collimator (FC). The SMF is also connected to a polarization rotator (PR) followed by a singlemode lensed-tapered fiber (LTF) designed for a 2.5 µm spot-size and working distance of 14 µm at λ = 1.55 µm (Oz optics). The LTF couples into a 5 µm × 220 nm fan-out tapered silicon waveguide, which adiabatically decreases in width over a length of 450 µm to a 520 nm × 220 nm channel waveguide. The PR is used to ensure that only the quasi-TE (QTE) mode of the waveguide is being excited, which produces maximum photocurrent as the overlap with the Zn + implanted region is larger than that of the quasi-TM mode. The channel waveguide adiabatically tapers into a rib/ridge waveguide with a 50 nm ridge and the same channel dimensions over a length of 100 µm, which then enters the Zn + implanted-silicon-waveguide photodiode. Reverse-bias voltage is applied and current is measured using a Keithley 2400 Source/Meter through electrical probes, which are landed on Al contact pads. Power transmitted through the photodiode is out-coupled in an identical manner to the input coupling, and passes through an SMF patch cable to a Yokogawa AQ6375 optical spectrum analyzer.

Determination of error bars:
The total fiber-to-fiber transmission loss is measured by aligning two lensed-tapered fibers at their focal distance, and is defined in dB as 2 = − , where and are defined in Fig. S1. This loss is averaged over four measurements to determine the mean and standard deviation, ̅ 2 and 2 , respectively. The transmission loss through the connecting fiber in dB is found as = − , and is also averaged over four measurements to find the mean and standard deviation ̅ , . The loss due to transmission through a single LTF is calculated as ̅ =

Estimation of Zn and defect concentrations after implantation:
The SRIM calculated dose profile is shown in Fig. S3. The averaged Zn concentration in the waveguide is estimated by integrating the Gaussian fit over the 220 nm length of the silicon device layer, and multiplying by the dose. The ion stopping range is adjusted by changing the acceleration voltage. For the PDs explored here, an acceleration voltage of 260 keV was used to place the stopping range in the center of the silicon device layer such that the fundamental quasi-TE mode of the SiWG (Fig 1b, main text) had maximum overlap with the created trap-states. The amorphization threshold is taken to be the dose for which the concentration of implantation-induced lattice defects accounts for 10% of the total available Si lattice sites. This corresponds to a defect concentration of 5 × 10 21 cm -3 for Si. Based on the SRIM calculations the threshold dose for amorphization is 2×10 13 cm -2 ; however, these calculations do not take thermal effects into account. Thus, the SRIM estimation is an underestimate of the amorphization threshold dose. Even so, this conservative estimate shows that the doses explored here are below the amorphization threshold. power could not be measured at these wavelengths due to the longer PD length, and thus error bars could not be determined for the responsivity.

Fig. S5 | a
Responsivity as a function of reverse bias voltage for 3 mm length PDs. Inset: dark current as a function of reverse bias voltage for an un-implanted diode (black dashed line) and PDs with Zn + doses of 10 12 cm -2 (red solid line) and 10 13 cm -2 (blue solid line). d Responsivity as a function of wavelength for 3 mm length PDs. Photocurrent was measurable above 2.35 µm; however, the transmitted power was not sufficient for determining error bars for the responsivity.