In this work, we demonstrate In0.52Al0.48As top/backside-illuminated avalanche photodiodes (APD) with dual multiplication layers for high-speed and wide dynamic range performances. Our fabricated top-illuminated APDs, with a partially depleted p-type In0.53Ga0.47As absorber layer and thin In0.52Al0.48As dual multiplication (M-) layer (60 and 88 nm), exhibit a wide optical-to-electrical bandwidth (16 GHz) with high responsivity (2.5 A/W) under strong light illumination (around 1 mW). The measured bias dependent 3-dB O-E bandwidth was pinned at 16 GHz without any serious degradation near the saturation current output. To further increase the speed, we downscaled the active diameter and adopted a back-side illuminated structure with flip-chip bonding for batter optical alignment tolerance. A significant improvement in maximum bandwidth was demonstrated (25 versus 18 GHz). On the other hand, we adopted a thick dual M-layer (200 and 300 nm) and 2 μm absorber layer in the APD design to circumvent the problem of serious bandwidth degradation under high gain (>100) and high-power operation which significantly enhanced the dynamic range. Due to dual M-layer, the carriers could be energized in the first M-layer then propagate to the second M-layer to trigger the avalanche process. In both cases, despite variation in thickness of the absorber and M-layer, the cascade avalanche process leads to values close to the ultra-high gain bandwidth product (GBP) of around 460 GHz with a responsivity of 0.4 and 1 A/W at unit gain for the thin and thick M-layer devices, respectively. We successfully achieved a good sensitivity of around −20.6 dBm optical modulation amplitude (OMA) at a data rate of 25.78 Gb/s, by packaging the fabricated APDs (thin dual M-layer (60 and 88 nm) version) with a 25 Gb/s trans-impedance amplifier in a 100 Gb/s ROSA package. The results show that, the incorporation of a dual multiplication (M) layer structure in the APD opens a new window to obtaining the higher GBP in order to meet the requirements for high-speed transmission without the need of further downscaling the multiplication layer.