The original 4096-bit Selectron was a large (5 inch by 3 inch) vacuum tube with a cathode running up the middle, surrounded by two separate sets of wires forming a cylindrical grid, a dielectric material outside of the grid, and finally a cylinder of metal conductor outside the dielectric, called the signal plate.
The two sets of grid wires were normally "biased" slightly negative, so that the electrons from the cathode could not flow through the grid and reach the dielectric. To select a bit, two adjacent wires on each of the two grids were biased positive, allowing current to move to (or from) the dielectric.
Writing was accomplished by selecting a bit, as above, and then sending a pulse of potential, either positive or negative, to the signal plate. With a bit selected, electrons would be pulled onto (with a positive potential) or pushed off of (negative potential) the dielectric. When the bias on the grid was dropped, the electrons were trapped on the dielectric as a spot of static electricity.
To read from the device the current direction was reversed. A bit was selected and a pulse sent into the cathode. If the dielectric for that bit contained a charge, the electrons would be pushed off the dielectric and read as a brief pulse of current in the signal plate. No such pulse meant that the dielectric must not have held a charge.
The smaller capacity 256-bit system was constructed similarily, but build in a planar fashion rather than cylindrical, resulting in an even larger vacuum tube. Each one cost about $500 to build, and while they were more reliable and faster than the Williams tube, their cost meant they were used only on a few machines. Both systems were replaced in the market with core memory, as soon as that became widely available.