Companion to core-memory.md. Scope is chapter VII of the 1965 book, book
pages 41 to 52, which is the element itself, the three basic operations, and
the cell catalogue. Chapter VIII onward, the nodes and the arithmetic unit, is
a separate job.
A controlled transformer of current pulses, figure 4. Toroidal ferrite core with a rectangular hysteresis loop, carrying a supply winding, a load winding and one or more control windings, with a diode in series with the load.
The parameters are chosen so that a supply pulse always drives the core into saturation regardless of where it started, and at the end of the pulse the core sits at point 0 on the loop. That single design decision is what makes the element digital: the supply pulse resets the state, and the control pulse decides whether the next reset does work.
pulse only produces edge transients, nothing transfers, and the supply generator sees almost no load.
amp-turns and set the core to point 1, saturation in the other direction. The next supply pulse then swings the whole loop, so the magnetising current is a small part of the pulse and the remainder transfers into the load.
The diode orientation is what separates the two directions: it passes the transformed supply pulse and blocks the EMF induced by the control pulse. The supply winding is fed from a high impedance source, so the control pulse cannot dump into it either. Control energy therefore goes only into flipping the core and can be far smaller than the supply energy delivered to the load. That ratio is the gain, and it is the reason a chain of elements works at all.
Threshold logic, implemented as algebraic summation of amp-turns. All control windings have identical turns and are always connected in series, so the current cannot branch and the amp-turns differ only in sign, which is set by the direction the winding is wound.
produces an output pulse.
suppresses the output.
A cell fires when the signed sum crosses the threshold. Everything else is wiring:
input. Where that is inconvenient, use separate positive windings on one amplifier.
connected to the other supply phase, sized to equal exactly one standard control winding. One input is then never enough and two coincident inputs are. The same one-turn winding connected positively instead of negatively is a unit generator, a continuous pulse train.
The book is specific about the conditions: the pulses must arrive simultaneously, the inhibit must last at least as long as the inhibited pulse, and the inhibiting amplifier must never be loaded more heavily than the inhibited one.
Two-phase supply. Phase one feeds amplifiers 1 and 3, phase two feeds 2 and 4, and the working half-cycle of one phase is the control half-cycle of the other, so one stage of delay is half a supply period. Phases overlap deliberately: while an amplifier is settling and emitting a transient on the front of its pulse, the next amplifier is still under drive from the other phase and is therefore insensitive to it. That overlap is not a refinement, it is the noise immunity of the entire machine.
Two amplifiers paired with mutual inhibition, the output of each feeding the inhibit winding of the other. A pulse on the upper input means +1, on the lower input means -1, neither means 0, and both at once also means 0 because they cancel.
Two details worth keeping. The signals at the input are bipolar even though the two polarities travel on separate amplifiers, because the negative polarity arrives as an induced EMF on the inhibit winding of the opposite core. And the mutual compensation is a deliberate noise measure: with no input, parasitic signals cancel in the pair instead of being amplified. That is the same symmetry as the memory cell, two cores per trit, and it is not a coincidence, it is one design idiom used twice.
Eighteen cell types on plastic boards, distinguished by amplifier count and by the function wired into the board. One type carries a transistor output, for driving the series-connected inhibit windings of several amplifiers in the case where, absent inhibition, two or more of them would fire at once. It is rated for up to eight simultaneously firing amplifiers. That is the fan-out limit stated honestly: because inhibit is a series current loop, fan-out is bounded by drive voltage and winding resistance, not by logic depth.
Output pulse duration depends on the load. The book gives the number: an amplifier driving a positive winding under inhibit runs nearly short-circuited and its pulse comes out 30 percent longer than when it is not inhibited. Hence two amplifier classes, simple and powerful, with the rule that a powerful amplifier must never be operated into a short circuit, because its output pulse would then be longer than the supply pulse itself.
This, not the logic, is what constrains a rebuild. Any redesign has to keep the loaded pulse shorter than the supply period, and that couples frequency, turns, core area and fan-out into one budget that cannot be optimised piecewise.
generator chapter. Drive it resonantly and recover the flux energy into a capacitor rather than dissipating it every cycle. Magnetic logic is lossy mainly because the drive energy is thrown away, and recovery is the single largest efficiency win available.
product of turns and area scales as 1/f. At 2 MHz, 52 turns becomes about 5, or the core shrinks tenfold, or a mix. But core loss rises, the pulse-width budget above tightens, and squareness degrades, so 1 MHz is a more credible target than 10.
tape cores give better squareness and work higher, which buys threshold margin, and margin is worth more here than raw flux density.
lower forward drop. This matters more than it looks, because the diode sets the load-side threshold, and diode spread was a large part of the matched pair sorting that made the cells expensive.
cells, and all of the cost is labour, sorting and tuning. Keeping the array magnetic and moving decode and sequencing into a replaceable module gives most of the durability with a fraction of the pain. The element deserves to be kept for inspectability and for having nothing to wear out, not for speed or density. It is slow and large by modern standards and always will be.
Four numbers from one cell on the bench decide the project:
If the margin is under roughly 20 percent, the cell is not viable without sorting, and sorting is precisely what made Setun's elements expensive. That is the go or no-go result, and it arrives from a dozen cores and an oscilloscope, before any machinery is bought.
appendices or chapter XIV may.
an argument.
breaking the pulse-width rule.
give a real bill of materials.
ref/setun-1965/ch07-logic.md, page markers included.destroys circuit drawings, so anything redrawn from them must be read off the page images.