We present studies on the limits of oxide reliability on a local, microscopic scale, using scanning tunneling microscope (STM)-based ballistic electron emission microscopy/spectroscopy (BEEM/S). In these studies, electrons are injected from the STM tip into the conduction band of a SiO2 layer that is imbedded in a metal-oxide-semiconductor (MOS) structure. The electron energy is determined both by the tip bias that can be set up to -13 V and by the applied oxide bias. Combining the two biases can heat electrons to energies that are unreachable in thin oxides by conventional Fowler-Nordheim injection methods. Our studies indicate that breakdowns are difficult to achieve for 7.1 nm oxides. A local breakdown was not observed even for an injected charge dosage of 1.8X10(3) C/cm(2) at equivalent Fowler-Nordheim stress fields of similar to 25 MV/cm, although defect densities in the oxide were as high as similar to 5X10(13)/cm(2). Evidence of anode hole injection is also observed under high oxide biases similar to 8 MV/cm. Therefore we conclude that trap creation and hole injection processes are not sufficient to cause breakdowns at arbitrary locations on the 7.1 nm oxides. Whereas electron trapping is dominant during electron injection for 7:1 nm oxides, only a positive charge buildup was observed in the 2.8 nm oxides while stressing with only 1 eV electrons. For 2.8 nm oxides, a local breakdown did not occur for dosages of 3.1X10(3) C/cm(2) at equivalent fields >43 MV/cm. The observed breakdowns were accompanied by gate metal failure and are hence believed to occur at weak spots in MOS capacitors. We conclude that an intrinsic breakdown limit of SiO2 has not yet been reached.