AlGaN/GaN heterostructures were regrown on semi-insulating GaN templates with varying threading dislocation density (TDD). Regrowths were carried out under Ga-rich conditions by plasma-assisted molecular beam epitaxy (PAMBE) to assess the correlation between threading dislocation density and gate leakage. It has been proposed that Ga-decorated threading dislocation cores in Ga-rich PAMBE-grown GaN may provide conductive channels for vertical leakage . Semi-insulating GaN templates with TDD of (i) ~2x10<SUP>10</SUP>, (ii) ~5x10<SUP>8</SUP>, and (iii) ~5x10<SUP>7</SUP> cm<SUP>-2</SUP> were selected, with regrown layers yielding the same TDD as that of the template. Template i was grown by Ga-rich PAMBE on a 4H-SiC substrate. Template ii was grown by MOCVD with Fe-doping on 4H-SiC, and template iii was grown by MOCVD with Fe-doping on a Lumilog free-standing GaN substrate. In order to provide low leakage buffer layers, C-doping was applied to compensate donor impurities at the regrowth interface. Growth of the GaN:C layer and subsequent GaN buffer were modulated to control Ga droplet accumulation. An AlGaN/GaN heterostructure was also grown by Ga-rich PAMBE directly on 4H-SiC for comparison with the regrown heterostructures. HRXRD data was used to determine the structural quality of the regrown films, as well as the thickness and Al content of the Al<SUB>x</SUB>Ga<SUB>1-x</SUB>N cap. The resulting heterostructures were Al<SUB>x</SUB>Ga<SUB>1-x</SUB>N/GaN with x=0.25±0.03 and Al<SUB>x</SUB>Ga<SUB>1-x</SUB>N thickness of 28±1 nm. AFM was used to characterize the surface morphology of the samples (Fig. 1). After etching the AlGaN layer between the source and drain, spaced 1 μm apart, buffer leakage measurements were performed. Drain biases to reach a buffer current of 1 mA/mm were 160-180 V, 55-65 V, 50-60 V, and 20-35 V for the direct growth, regrowth on template i, regrowth on template ii, and regrowth on template iii, respectively. Gate leakage measurements (Fig. 2) were performed on large-area Schottky diodes for all samples. Decreasing TDD from ~2x10<SUP>10</SUP> to ~5x10<SUP>7</SUP> cm<SUP>-2</SUP> yielded up to a 45-fold decrease in mean gate leakage at -10 V bias. Passivated recess gate HEMT structures were fabricated and subjected to three-terminal breakdown measurements biased 1 V below pinch-off. To reach 1 mA/mm, drain biases of 15-20 V, 20-25 V, and 30-33 V were applied to the direct growth, regrowth on template i, and regrowth on template ii, respectively. The breakdown voltage of template iii was limited by buffer leakage. An AlGaN/GaN heterostructure has been regrown on template iii with a lightly C-doped buffer layer to reduce buffer leakage.