How to produce infrared optoelectronic silicon
Silicon is the material of choice for optoelectronic technologies ranging from the imagers in camera phones to photovoltaic cells, because of its abundance, low cost, existing manufacturing infrastructure, and bandgap of 1.1eV, which makes silicon conducive to collection of visible and near-infrared photons. Photons with lower energy, however, pass through silicon unabsorbed (see Figure 1). As a result, silicon solar cells fail to capture a significant portion of the solar infrared (IR) spectrum, and detectors and imagers operating at wavelengths greater than 1100nm must be fabricated from materials that are expensive and challenging to integrate with silicon. If silicon could capture this light, solar cells could be more efficient, and IR optoelectronic devices could be fabricated at considerably lower cost. Introducing mid-gap states (for example, by doping the silicon with selected impurities) permits these photons to be absorbed. As more impurity is added, the individual states broaden into a band, potentially giving access to an intermediate band effect (two-photon absorption) or to sub-gap photoconductivity (see Figure 1). Over the last decade, researchers have been working to capitalize on the observation that silicon strongly absorbs IR when it is heavily doped with chalcogens.1, 2
This can be achieved by ‘laser hyperdoping.’ When a nanosecond-pulse laser induces melting in a silicon layer heavily ion-implanted with an impurity, the rapid solidification that follows has two important benefits. First, if a sufficiently thick layer of silicon melts through the layer damaged by the implantation process to the perfect crystal underneath, the resultant layer will grow epitaxially from the underlying layer. Second, if the pulse is short enough, the impurity will become kinetically trapped in the solid, in concentrations that can exceed the equilibrium solubility limit by orders of magnitude. The resulting ‘hyperdoped’ layer is of great interest for IR optoelectronic devices.3, 4
However, it has been shown that at high chalcogen concentrations, the impurity band overlaps the conduction band, negating the possibility of sub-gap photoresponse. In other words, all of the carriers that are promoted to higher energies by absorbed photons were already above the Fermi level, and therefore do not contribute to a change in the conductivity of the layer.5 Such a layer will exhibit high sub-gap absorption, but negligible sub-gap device external quantum efficiency (EQE), that is, the ratio of the number of photogenerated carriers to the number of photons incident on the device (see Figure 1).
Several possible strategies have been proposed to solve this. The most obvious is simply to select an impurity whose states sit further into the middle of the silicon band gap, so that its broadened band would not overlap the silicon conduction band. While this sounds simple in principle, nearly all of the candidate impurities that meet this criterion are transition metals. These turn out to be challenging to incorporate into silicon in appreciable concentrations even by laser hyperdoping, because they have interface diffusivities that are orders of magnitude higher compared to conventional dopants. In other words, they can more easily escape the solid-liquid interface without becoming kinetically trapped in the solid.
A quantity that captures the difficulty for incorporation of a given impurity is the impurity's ‘diffusive speed,’ the ratio of the interface diffusivity to the interface width. Conventional silicon dopants such as arsenic and boron have diffusive speeds of the order of 100–101m/s. (Chalcogens fall at the low end of this range.) In contrast, transition metals have diffusive speeds of the order of 102–104m/s.6 In practical terms, this means that higher solidification speeds are needed for kinetic trapping of these impurities. Neodymium-doped yttrium aluminum garnet (Nd:YAG) lasers treble the solidification speed compared with excimer laser melting, and this has enabled trapping of gold at concentrations greater than 1019atoms/cm3 in a crystalline layer 150nm thick.
When a simple device is made out of such a layer (see Figure 2 for an illustration), IR device response at wavelengths as long as 2200nm is observed.7 Figure 2 shows a map of the EQE at 1550nm, obtained by rastering (scanning in a series of lines) a focused 1550nm laser over the surface and measuring the response under bias. The EQE is as high as 10−4, although it must be admitted this is still rather low for applications. One particular deleterious consequence of the diffusive speed is the limit that it imposes on the thickness of the gold-doped layer. Thus, while the layer exhibits an optical absorption coefficient greater than that of germanium at 1550nm, the total amount of light absorbed is low because the layer is so thin. Recent research should enhance sub-gap absorption and, correspondingly, the device EQE.7 Although this is an early result, it holds promise for the efficacy of transition metal hyperdoping as a strategy for fabricating silicon devices with usable photoresponse in the IR. We are now working to improve the EQE, which could be achieved in several ways, including by increasing the gold dose, altering the device architecture, or implementing light trapping.
Jeffrey Warrender received his PhD from Harvard University. His current research involves laser interactions with materials.