Semiconductor lasers are an integral part of high-speed telecommunications. The push for higher modulation frequencies, thereby allowing greater data rates, has motivated the scientific community for several decades. However, the maximum speed of directly-modulated semiconductor lasers has plateaued as the field reaches a mature state. Recently, optical injection locking has been proven to enhance the bandwidth and resonance frequency of directly-modulated semiconductor lasers. The injection locking technique allows the lasers to exceed their fundamental modulation speed limit, allowing for greater communication speeds. However, although the resonance frequency has been predictably linked to the injection locking parameters, the bandwidth enhancement has not been reliably correlated to the resonance frequency, unlike typical directly-modulated lasers.

In this dissertation, we first develop theoretical insight into the nature of resonance frequency and bandwidth enhancement, attempting to correlate the two. We describe the fundamental limit of resonance frequency enhancement and generalize these results to oscillators of all kinds. Using these theoretical trends, we optimized the injection locking performance of 1550 nm distributed feedback lasers. We report a high-speed resonance frequency of 72 GHz and a 3-dB modulation bandwidth of 44 GHz. These are the highest reported resonance frequency and 3-dB bandwidth of any directly-modulated semiconductor laser, respectively.

Direct measurement of laser frequency response is often limited by the bandwidth of photodetectors and network analyzers. In order to measure frequencies above our detection equipment limit (50 GHz), we develop a new optical heterodyne technique that can detect arbitrarily-high modulation frequencies. This technique, in contrast to previous heterodyne methods, does not require stable frequency solid-state lasers and can be used to test telecom-wavelength lasers.

Finally, we discuss a new modulation technique, where the master is modulated rather than the slave. This technique has many applications, such as residual amplitude modulation reduction, frequency modulation regeneration, and frequency discrimination. We demonstrate the latter experimentally, achieving 0.88 mW/GHz frequency-to-amplitude conversion. Additionally, we develop the basis for the theory that governs these techniques and find the theory in good agreement with our experiments.