previous | contents | next

DESIGN DECISIONS FOR THE PDP-11/60 MID-RANGE MINICOMPUTERS 317

velopments. Memory is the most basic component of a computer, and it is utilized throughout the design. In addition to obvious uses as main program and data memory, and as file storage devices (disks and tapes), memory is also located within the central processor in the form of registers, state indicators, control, and buffer storage between the central processor and main (primary) memory. In input/output (I/O) devices, there are buffers and staging areas. Memory can be substituted for nearly all logic by substituting table lookup for computation.

The constantly increasing bit density mentioned previously has been the most dramatic development in memories. For example, bipolar read-write or random-access memory (RAM) chips have advanced as follows.

Cost reductions have paralleled bit density increases. A consequence of high density RAM technology is that cache memories are now extensively used in mid- and upper-range minicomputers. Bipolar ROM densities have led RAM densities by about a year. Thus, the 2048- bit ROM, organized as 512X4, was available in 1975.

These factors have made microprogrammed control increasingly attractive to the mini computer designer. While large-scale computers utilized extensive microprogramming during the 1960s, it was not a cost-effective choice for the minicomputer designer because of the prohibitive cost of the read-only storage technology then available.

Both hardwired control devices and microprogrammed control devices have curves that trace increases in cost as they implement increasing functionality (Figure 3). However, the rate of cost increase is less for micro programmed controls than for hardwired controls. Davidow [1972] demonstrates that a factor of 4 difference exists between the two slopes.

At some point, the two related hardwired and microprogrammed curves cross. Beyond that intersection, microprogrammed controls are

Figure 3. Semiconductor technology trends in control implementations. Cost comparisons, at three different points in time, of conventional hardwired control and advanced microprogrammed control show two important trends. First, at fixed point in time in 1 970s (e.g., time t3), microprogrammed control is less expensive above certain level of complexity (x3). For simplest type of ma chine, random logic gives most economical design. Microprogrammed design has base cost associated with address sequencing and memory selection circuitry. Microprogrammed control cost increases slowly with number of sequencing cycles, which are added as complexity increases, because each additional cycle requires one additional word of control store. Second, because rate of cost-decrease for memories is greater than the rate for random logic, crossover points move with time, gradually shifting in favor of microprogrammed control. When 11/20 was designed (time t1) hardwired controls were cheaper. Its successor, the 11/40, was designed at time t2 and used microprogramming. The 11/60, at time t3, used increased microprogramming.

previous | contents | next