Fabrication of diffused diodes for HP lightwave applications - Technical

Hewlett-Packard Journal, Feb, 1993 by Patricia A. Beck

The simple but robust p-i-n dual detector used in the receiver of the HP 8504A precision reflectometer has -17 dB return loss (2% reflection) operating at both 1300 nm and 1550 nm.

The HP 8504A precision lightwave reflectometer uses a pair of cofabricated planar photodiodes as its receiver element (see article on page 39). One path of the split polarization beam strikes each diode and the resulting photocurrents are added vectorially. Although the InP/InGaAs detectors have been optimized for wavelengths from 1300 nm to 1550 nm, other operational configurations are possible.

P-i-n photodetectors convert optical inputs to electrical signals. A photon enters through the p-doped region, which is chosen to be transparent to the radiation, and is absorbed in the intrinsic (i) region. An electron-hole pair is formed. Carriers are moved from the intrinsic region to the p and n regions by the electric field induced by the applied voltage bias. The movement of charge is the detected photocurrent. This photodetector operation is shown in Fig. 1.

This article describes the design and process used to develop the diffused photodiodes used in the HP 8504A.

Types of Photodiodes

Two types of semiconductor photodiodes are used in HP lightwave products: mesa and diffused planar. While their operation is similar, their structures are different.

Mesa p-i-n diodes have their layers grown epitaxially with the dopants already in place. Extraneous material is then etched away to form the devices and provide isolation and separation from neighboring devices as shown in Fig. 2. The layers for the devices in question are: Zn-doped/nP for the p region, sulphur-doped InP for the n region, and undoped (intrinsic) InGaAs for the i region. Planar p-i-n diodes also have their layers grown epitaxially, but the layers above the n-type substrate are undoped. The p region is formed later by diffusing dopants into the surface. After the epitaxial layers are grown, a diffusion barrier (in this case silicon nitride) is deposited and patterned. The openings in the diffusion barrier expose the InP surface to define and separate devices. The patterned wafer is sealed in a closed vessel with the solid dopants or placed in an OMVPE (organometallic vapor phase epitaxy ) reactor with dopant gases flowing over its surface. This may be the same type of reactor used to grow the original epitaxial structure. Fig. 3 depicts a wafer resting in a quartz ampoule with elemental zinc (the p dopant), arsenic (to slow and control the zinc diffusion), and phosphorous (to provide an overpressure to keep the phosphorous in the InP from diffusing out at high temperatures). Out-diffusion causes severe surface pitting. The sliding bar in Fig. 3 illustrates the balance of Zn, As, and P in the ampoule. Holding the As concentrate constant while increasing the amount of Zn tends to cause a poor, pitted surface (100% Zn with 0% P is the extreme). At the other end of the scale, 100 degrees /5 P and 0% Zn will result in a smooth surface but no diffusion. The exact balance of elements depends on the process parameters for a specific device. The wafer is held at high temperature long enough for the solid sources to form a gas and drive the dopants to the InP/InGaAs junction. Although the time is more predictable in gas source diffusion because there is no solid phase with which to contend, closed ampoule diffusion is adequate for small production runs and is used in this process. Diffused photodiodes can easily be sized from micrometers to millimeters, and can be closely spaced. The planar design offers easier fabrication and passivation with improved leakage characteristics and an abrupt turn-on. The particular birefringent crystal chosen for use in the HP 8504A splits the incident radiation into two beams whose polarizations are orthogonal. Two planar detectors fabricated on the same microchip offer advantages of matched spacing (to the beams) and matched responsivity (to each other). Fig. 4 shows the dual detector used in the HP 8504A. Bond pads lie off the active area above the diffusion barrier.

Processing and Development

To speed processing (both layout and alignment) and to increase yield, a simple tab contact to one side of the active region is used rather than an encircling ring of metal. If the diode is probed with a small light beam, response decreases slightly with distance from the contact. In the HP 8504A, the beam striking the surface is large compared to the diode, therefore, the tab design is adequate. Although it has higher capacitance and therefore lower frequency response than a mesa device with a similar intrinsic layer thickness, the diffused device is ideal for this broad-area application. Under reverse bias conditions the intrinsic layer of the diode is depleted of charge. In the absence of light only a small reverse leakage or dark current (Id) flows. Id is considered a measure of the noise floor of the receiver. Planar devices have lower dark currents than mesa devices. A typical 80-micro-m-diameter mesa device from this laboratory exhibits Id less than 30 nA at -5V, while a similarly sized planar device would have an Id of less than 1 nA. Dynamic resistance (extrapolated resistance through OV) is greater than 200 M[Ohms] Responsivity is on the order of 0.9 to 1.1 A/W over the wavelengths of interest between OV and -5V bias. Fig. 5 graphs typical current-versus-voltage characteristics.