Since the 1940s a steady stream of electronic devices for producing randomness have been invented, with no single one ever becoming widely used. An early example was the ERNIE machine from 1957 for British national lottery (premium bond) drawings, which worked by sampling shot noise from neon discharge tubes—and perhaps because it extracted only a few digits per second no deviations from randomness in its output were found. (U.S. missiles apparently used a similar method to produce randomly spaced radar pulses for determining altitude.) Since the 1970s electronic randomness generators have typically been based on features of semiconductor devices—sometimes thermal noise, but more often breakdown, often in back-biased zener diodes. All sorts of schemes have been invented for getting unbiased output from such systems, and acceptable randomness can often be obtained at kilohertz rates, but obvious correlations almost always appear at higher rates. Macroscopic thermal diffusion undoubtedly underestimates the time for good microscopic randomization. For in addition to 1/f noise effects, solitons and other collective lattice effects presumably lead to power-law decay of correlations. It still seems likely however that some general inequalities should exist between the rate and quality of randomness that can be extracted from a system with particular thermodynamic properties.