The development of semiconductor products increasingly requires doing high-speed measurements with internal probe. For high-speed DRAMs, this demands a combination of sub-20 fF capacitance, sub- 100ps sampling time, and sub-20ps accuracy that is not available with commercial mechanical probes. Commercial e-beam probers are the most promising tool, but their suitability for accurate measurements has not been determined. This paper reports the timing accuracy for the IDS-10K+ prober to establish the basis for quantitative high-speed measurements. The ultimate accuracy of the IDS-10K+ is determined by inherent errors due to timebase inaccuracy and voltage noise. These factors yield a uncertainty of ±. 60ps when measuring time intervals of 500 ps or longer, ± 10% for intervals in the range 80-500ps, and a minimum error of ± 8ps for intervals shorter than 80ps. Timebase drift causes an additional uncertainty of less than ± 20 ps over a 4 hour period. Internal signals were measured on DRAM circuits to determine the effect of all errors. Simple methods are shown for measuring sub-100ps edges and for determining the sampling time. The minimum sampling time is shown to be 65ps for a 40ps pulsewidth. Edge positions were repeatedly measured to quickly determine the uncertainty from all random errors. For sub-500ps edges, this is less than ± 10 ps for metal probe points and ± 16 ps for passivated circuits. For edges with transition times longer than 500ps, the uncertainty increases with transition time. This is consistent with the observed distortion of these waveforms by a changing low-frequency background, so that timing accuracy becomes dominated by voltage errors for slow edges. Mechanical probe and e-beam measurements of delay times were compared for a set of 9 signals. The observed 30-400 ps discrepancy cannot be explained by the effect of the mechanical probe's risetime or load capacitance. The largest disagreement occurs in a particular region of the circuit and the discrepancy increases with transition time. It appears that local interference is the dominant source of errors that are not inherent to the prober. When the interference is synchronous with the test cycle, systemmatic errors result. When the interference is asynchronous, random errors are produced. FIB probe points were made on a passivated sample to compare delays times measured on bare metal and on the passivation surface. Edges probed on the passivation were shifted by ± 20 ps with respect to bare metal. The sources of error are now sufficiently characterized to use the IDS-10K+ for valid high-speed internal probe, in which the uncertainty for any given measurement is already known or readily determined.