3 ms·

> and why each ring has three wires woven through it (and I still don’t understand exactly how these work). The rings are made from hard-magnetic ferrite. Basi

> and why each ring has three wires woven through it (and I still don’t understand exactly how these work). The rings are made from hard-magnetic ferrite. Basically permanent magnets. To write these, a positive or negative current pulse is applied to a write line. This will induce a voltage in the read line. The magnitude of the induced voltage depends on whether the write pulse coincides with the remanent field (the polarity of the ring). This allows you to store one bit. Note how reading and writing are the same thing, except that when reading you always use e.g. a positive current and then re-write the bit immediately after. So writing = single cycle operation. Reading = destructive and hence two cycles. This is inefficient though, because for each bit you need two wires. Genius idea: Let's split the write line up into two write lines, each carrying half the write current. Lay out cores in a matrix and you can address individual cores, because only the core where both wires cross will reach the field strength necessary to be re/de-magnetized. The third wire is the read line, which is looped through all cores. This works because only a single bit is read at a time.

jacquesm 8y ago
> Basically permanent magnets. Basically iron, which can be magnetized or demagnetized just like you can use a magnet to magnetize a steel needle.
kens 8y ago
Since I've been maintaining core memory recently, let me add a few comments. Core memory is built from ferrite rings called cores, which can be magnetized in two different directions. By passing a current through a core one way, it can be magnetized the matching direction, and passing a current the other way magnetizes the core the other direction. The important property for reading a core is that when a core flips from one state to the other, the changing magnetic field will induce a voltage on the sense line threaded through all the cores. But if the core stays the same, you don't get a voltage. Thus, by flipping a core to 0, you can detect if the core was in the 1 state previously. This allows you to read a core, but the read is destructive since the core is now 0. Another key property of the cores is hysteresis. A small current through a core has no effect on it, but above a threshold the current can flip the core. This is very important because it means that you can put cores in a grid, and energize one X line and one Y line. Only where the two lines pass through the same core will the core receive enough current to potentially flip. This coincident-core technique is what made core memory practical. This grid of cores is called a core plane. For a 16-bit computer, you stack up 16 core planes and read or write them in parallel. Finally, to write a core, you first access it to ensure it is flipped to 0. Next you access it with currents in the opposite direction, flipping it back to 1. But what about the bits in the word that you don't want flipped to 1? The trick is the inhibit line, which passes through all the cores in the a, in the opposite direction. If you put a current through the inhibit line in a plane at the same time as the write, the currents partially cancel out, and the core doesn't flip, so it stays as 0. Thus, each core typically has 4 wires: X and Y wires in a grid to select the core, a sense line that passes through all the cores in a plane to read the value, and an inhibit line that passes through all the cores in a plane to inhibit the writing of a 1. Some systems combine the sense and inhibit lines into one line, but that increases noise. (You want the sense line to switch directions so noise from the X and Y lines cancels out (that's why it's usually diagonal). But the inhibit line is forced to run a particular direction so the current cancels out.) If this isn't more than you ever wanted to know about core, I've written about the core memory in the IBM 1401 in detail here: http://www.righto.com/2015/08/examining-core-memory-module-inside.html http://www.righto.com/2015/08/examining-core-memory-module-i...
- jmiserez 8y ago
  Great explanation and article. Discussion (2015) here: https://news.ycombinator.com/item?id=10143700 https://news.ycombinator.com/item?id=10143700

jacquesm 8y ago

> Basically permanent magnets. Basically iron, which can be magnetized or demagnetized just like you can use a magnet to magnetize a steel needle.

kens 8y ago

Since I've been maintaining core memory recently, let me add a few comments. Core memory is built from ferrite rings called cores, which can be magnetized in two different directions. By passing a current through a core one way, it can be magnetized the matching direction, and passing a current the other way magnetizes the core the other direction. The important property for reading a core is that when a core flips from one state to the other, the changing magnetic field will induce a voltage on the sense line threaded through all the cores. But if the core stays the same, you don't get a voltage. Thus, by flipping a core to 0, you can detect if the core was in the 1 state previously. This allows you to read a core, but the read is destructive since the core is now 0. Another key property of the cores is hysteresis. A small current through a core has no effect on it, but above a threshold the current can flip the core. This is very important because it means that you can put cores in a grid, and energize one X line and one Y line. Only where the two lines pass through the same core will the core receive enough current to potentially flip. This coincident-core technique is what made core memory practical. This grid of cores is called a core plane. For a 16-bit computer, you stack up 16 core planes and read or write them in parallel. Finally, to write a core, you first access it to ensure it is flipped to 0. Next you access it with currents in the opposite direction, flipping it back to 1. But what about the bits in the word that you don't want flipped to 1? The trick is the inhibit line, which passes through all the cores in the a, in the opposite direction. If you put a current through the inhibit line in a plane at the same time as the write, the currents partially cancel out, and the core doesn't flip, so it stays as 0. Thus, each core typically has 4 wires: X and Y wires in a grid to select the core, a sense line that passes through all the cores in a plane to read the value, and an inhibit line that passes through all the cores in a plane to inhibit the writing of a 1. Some systems combine the sense and inhibit lines into one line, but that increases noise. (You want the sense line to switch directions so noise from the X and Y lines cancels out (that's why it's usually diagonal). But the inhibit line is forced to run a particular direction so the current cancels out.) If this isn't more than you ever wanted to know about core, I've written about the core memory in the IBM 1401 in detail here: http://www.righto.com/2015/08/examining-core-memory-module-inside.html http://www.righto.com/2015/08/examining-core-memory-module-i...

jmiserez 8y ago
Great explanation and article. Discussion (2015) here: https://news.ycombinator.com/item?id=10143700 https://news.ycombinator.com/item?id=10143700