M410_8c 194..197

194 NATURE | VOL 410 | 8 MARCH 2001 | www.nature.com ture. The integrated PL also increases with temperature in the same manner as the EL, indicating that the temperature dependence is intrinsic to the recombination, rather than the injection mechanism. The strong temperature quenching of PL and EL in most semiconductors is primarily the result of the strong temperature dependence of the competing non-radiative routes—the band-toband transition is relatively temperature-independent. In our case spatial localization of the radiative carrier population decouples it from any non-radiative recombination occurring elsewhere, thus eliminating the luminescence quenching. A full understanding of the form of the weak increase of the integrated intensity with temperature seen requires further investigation; it could be associated with increased scattering within the confined carrier populations and the increase in the effective density of band states with temperature. We have also carried out EL frequency-resolved measurements that give a device response time, at room temperature, of 18 6 2 ms. The PL and EL are both superlinear with excitation power and drive current respectively, meaning that the device improves further as we drive the device harder. In ref. 9 a similar dependence of the luminescence intensity is attributed, owing to a shift in emission wavelength, to sample heating. However, in our case, no such shift is observed. The external quantum efficiency of our unpackaged planar device has been measured, based only on the light emitted just through the back window. At a 100 mA forward current the emitted light is 19.8 mW giving an external quantum efficiency of ð2:0 6 0:1Þ 3 10 2 4 at room temperature. We also obtain significant edge emission from the device, of a further 80 mW, but this is not included in our estimate of the quantum efficiency above. Taking it into account, our device has a quantum efficiency of 10. The emitted power from the face of the device was measured by placing it immediately adjacent to a large-area calibrated power meter. The power meter is an Ophir Laser Star fitted with a PD300 IR head; the instrument is certificated and calibrated with standards traceable to the National Institute of Standards. Our front emitting devices have the same efficiencies as the back emitting devices. For comparison, commercially available GaAs infrared LEDs have typical external efficiencies of 10. However, commercial packing of LEDs, in particular by minimizing the considerable internal reflection losses of the planar device and by collecting the edge emission, can improve the external efficiency by up to a factor of about 15. Our first, non-optimized devices are therefore already within a factor of 3 or so of the efficiencies achieved in conventional optimized LED devices. We believe that the device demonstrated is the most likely candidate currently for implementing efficient light sources in silicon. Such a device would also form the basis for the development of an injection laser based on the same principles but with the incorporation of an optical cavity. The approach itself is not limited to silicon but could be applied to other materials, particularly other indirect materials and silicon alloys. For example, going from germanium to silicon to silicon carbide could produce devices, using the same approach, that could span the near-infrared region, including the 1.3 mm and 1.5 mm wavelength regions of the spectrum, important for optical fibre communications, and up to the ultraviolet region. M