hunt@spar.SPAR.SLB.COM (Neil Hunt) (07/21/87)
People have explained how two varieties of CRT memory worked, one based upon storage tube technology (flood guns, etc) and one based upon simple, non storage CRTs. I am intrigued by the problem of writing a zero (or unwriting a charged spot, at least) to the earlier, non storage tube type systems. I asked my colleagues, and we found a book called "IBM's early computers" by Bashe, Johnson, Palmer and Pugh. We found an explanation differing substantially from any we had heard (except Nigel Topham's notes about writing dots and dashes); let me quote: With the standard CRT as a point of departure, Williams turned to the use of secondary emission, a phenomenon unrelated to lumi- nescence, for information storage. The phosphor, when struck by a beam of electrons of sufficient energy, emits more than one secondary electron per impinging electron. It is thus possible to create a "well", a very small region of positive charge (due to loss of secondary electrons) at a selected spot on the tube face by directing the beam to that spot. It is possible to fill the well by directing the beam to a second spot nearby, typically just over one beam diameter away. Secondary elec- trons released from the second spot are attracted to the first by its positive potential and fill the well. The first spot of the pair is either charged positively (amd may be arbitrarily said to store a 0) or discharged (storing a 1) depending on whether its companion spot is also struck by the beam. The beam is directed to the second spot only if a 1 is to be stored. [...] The sudden changes in potential that occur at the face of the tube when the beam strikes the phosphor are of opposite sign, depending on whether the beam is "digging" a well or landing in one. These changes can be detected by an electronic amplifier connected to a conducting plate placed close to the exterior surface of the tube face. The value of a bit previously stored can be sensed by detecting the positive or negative signal (a stored 1 ort 0 respectively) that results when the beam is directed to the first spot of of a pair. If the signal pickup plate consists of a piece of wire screen or conducting glass fastened to the face of the CRT, the sotred information can be read visually by interpreting a single spot of light as a 0 and a pair of spots as 1. Williams, as well as others who experimented with his system, tried numerous variations on a pair of spots. In one version [...] the beam was merely deflected to the second spot while still turned on, if a one was to be stored. The result was a dot or dash display for each stored 0 or 1 respectively. The book goes on to present a diagram taken from Williams' and Kilburn's patent (2 777 971), and discusses various complications and other aspects. It claims that the tube was operated in random access mode, not in raster order; cycle time was around 20uS per bit. Such a system needs regeneration (refresh) for two reasons. First there is the fact that the charges leak away over time, but also there is the "spill" effect, where accesses to cells tend to partially fill neighbouring cells with surplus charge. A "read-around ratio" of 200 was a typical maximum accesses any hot spot cell could receive before the tube would require a regeneration cycle, in addition to the time requirement of a regeneration every few tenths of a second. Finally: As it turned out, the memory was one of the most troublesome components of the [IBM] TPM [Tape Processing Machine, circa 1951] system, if not the most troublesome. Its erratic behaviour was masked to a considerable degree, however, by a variety of problems throughout the system that stemmed from the experimental nature of many of its vital parts. Times change ! I wonder what article 1234512345 of comp.arch in 2024 will be discussing - those archaic silicon charge storage memories ? Neil/.