Ken Shirriff, well known for his vintage computer analyses, has an interesting article on the technology used by IBM System/360 mainframes for storing microcode (the kind of code which instructs the inner sequencing of the CPU) well before there was semiconductor memory. Specifically it’s about Transformer Read-Only Storage (TROS, IBM always used the term storage as in read-only storage or ROS, not memory as in ROM), consisting of a series of transformers crossing through a stack of Mylar sheets, encoding bits by either passing signal lines around or through the individual transformers. In principle this is not dissimilar from the famous rope memory of the Apollo Guidance Computer. In this case, however, the programming wasn’t done by passing wires, but rather by punching holes in the Mylar sheets to break unwanted signal paths. Think, rope memory meets punchcards. (There was also a predecessor technology using actual punchcard formats, CCROS, also described in the article.)
Ken Shirriff, TROS: How IBM mainframes stored microcode in transformers
Related HN discussion: https://news.ycombinator.com/item?id=21548757