The Intel 80386 microcode ROM contains 94,720 bits, significantly larger than the 10,752-bit microcode ROM of the 8086 processor [1]. This size difference made understanding the 80386 microcode far more complex.
Ken Shirriff provided a high-resolution image of the 80386 microcode ROM die, enabling detailed analysis [1]. Initially, disassembling the 80386 microcode posed a challenge because it appeared as a large undifferentiated binary blob. Unlike the 8086, no official documentation or patents existed to explain its structure [1].
A team featuring the researchers GloriousCow and Smartest Blob applied advanced computer vision techniques, AI algorithms, and human-guided automation tools to extract the binary microcode from the die image [1]. This approach allowed them to decode the microcode without official reference materials.
Through this process, the researchers identified recurring patterns showing that the microcode consists of micro-operations (μ-ops) with distinct fields representing source and destination registers as well as ALU inputs [1]. Ken Shirriff contributed by tracing connections in the die logic, which helped clarify the organization and functioning of the microcode [1].
The work reveals how modern AI and image processing can unlock legacy hardware secrets once thought inaccessible. The findings provide a detailed, machine-readable representation of the 80386 microcode. No further dates or next steps were given.