In this work, a fast 32-bit one-million-channel time interval spectrometer is proposed based on field programmable gate arrays (FPGAs). The time resolution is adjustable down to 3.33 ns (= T, the digitization/discretization period) based on a prototype system hardware. The system is capable to collect billions of time interval data arranged in one million timing channels. This huge number of channels makes it an ideal measuring tool for very short to very long time intervals of nuclear particle detection systems. The data are stored and updated in a built-in SRAM memory during the measuring process, and then transferred to the computer. Two time-to-digital converters (TDCs) working in parallel are implemented in the design to immune the system against loss of the first short time interval events (namely below 10 ns considering the tests performed on the prototype hardware platform of the system). Additionally, the theory of multiple count loss effect is investigated analytically. Using the Monte Carlo method, losses of counts up to 100 million events per second (Meps) are calculated and the effective system dead time is estimated by curve fitting of a non-extendable dead time model to the results (τNE = 2.26 ns). An important dead time effect on a measured random process is the distortion on the time spectrum; using the Monte Carlo method this effect is also studied. The uncertainty of the system is analysed experimentally. The standard deviation of the system is estimated as ± 36.6 × T (T = 3.33 ns) for a one-second time interval test signal (300 million T in the time interval).