Photoelectrochemical (PEC) water splitting is a sustainable process for generating hydrogen gas that can be stored efficiently and converted back to usable electricity in a fuel cell. Solar-to-hydrogen (STH) efficiencies in the range of 10-12% have been demonstrated using GaInP/GaAs dual junction photovoltaic-based devices immersed in electrolyte. With bandgaps of ~1.82 and 1.42 eV, respectively, these baseline devices generate an oversupply of photovoltage, even accounting for the kinetic overpotentials at both the HER and OER electrodes. Progress toward higher efficiency was demonstrated at NREL by growing a GaInP/GaInAs tandem with bandgaps of 1.82 and ~1.2eV, increasing the lattice constant of the lower bandgap material in order to access a lower bandgap alloy. PEC devices based on the “inverted metamorphic multijunction” (IMM) architecture demonstrated STH efficiencies >16%. However, because the very thin epitaxial films of the IMM are removed from the growth substrate and bonded to a secondary, mechanical handle, the IMM is susceptible to growth-related defects that lead to long-term stability problems. The specifics of the secondary bonding have also restricted the post-growth deposition temperature of protective coatings and co-catalysts to <140°C, which limits the available strategies for improving durability and efficiency. Notwithstanding the lower STH efficiency, the on-substrate GaInP/GaAs architecture may nonetheless have advantages in terms of stability, durability and processability that are potentially more important for a working PEC product than STH efficiency. Here we show that the baseline device can be improved by extending the wavelength range of photon absorption by including quantum wells (QWs) in the current-limiting bottom cell. By adding GaAsP/GaInAs QWs in the depleted region of the GaAs bottom junction, the absorption range can be extended from ~870 nm to ~930 nm, leading to an overall increase of >2.5 mA/cm2that is split between the two cells, or equivalently a >1.25 mA/cm2gain in the light-limiting photocurrent of a current-balanced tandem. This represents a >10% increase in the relative STH efficiency compared to a baseline GaInP/GaAs device. Since the baseline device has a voltage oversupply, the ~100 mV drop in voltage because of the QWs is unimportant. Quantum wells are composed of alternating barrier and well layers. Though the design is flexible and may be tuned to a different absorption edge, the design shown here is composed of GaAs0.9P0.1with a bandgap of ~1.55 eV, and Ga0.9In0.1As with a bandgap of ~1.27 eV. The thickness of the well layer is ~85 Å and the two-dimensional quantum confinement pushes up the effective band edge to 1.34 eV (930 nm). Although neither of these alloys alone are lattice-matched to the GaAs substrate, the entire QW region is carefully strain-balanced so that the net strain is nearly zero. The devices reported here were grown by atmospheric pressure metalorganic vapor phase epitaxy (MOVPE). Growth, fabrication and characterization of the devices will be discussed in greater detail at the conference. Figure 1a shows a not-to-scale schematic of the device with the main layers and QWs indicated. The two cells are separated by a tunnel junction interconnect. Figure 1b shows an (004) x-ray rocking curve of a test structure with 80 quantum wells. The narrow, high intensity peaks indicate good quality interfaces between the barrier and well layers. Strain-balancing these two alloys requires different thicknesses in the two layers, which gives rise to the asymmetry in the satellite positions. Figures 1c and 1d show the incident photon-to-current efficiency (IPCE) and current-voltage curves for a baseline GaInP/GaAs device and an 80-QW device. The increased absorption edge of the bottom cell is clearly visible in Figure 1c. Though not optimized, the increased light-limited photocurrent is visible at 0 V in Figure 1d. With a thicker top cell and improvements to the device isolation, an STH efficiency >13% is achievable with this architecture, even without an anti-reflection coating. Figure 1