In-Memory Computing

Processing data where it is stored rather than shuttling between separate memory and compute units — the architectural principle of neuromorphic systems.

The von Neumann bottleneck — limited bandwidth between processor and memory — dominates conventional computing’s energy consumption. In-memory computing eliminates this bottleneck by co-locating storage and computation, as brains do with synapses.

Systems Connection

In-memory computing is a structural innovation: the arrangement of components (where memory and logic are placed) determines function (computational efficiency). Memristor crossbar arrays exemplify this: the physical structure that stores weights also performs matrix multiplication.

Key Technologies

• Memristor crossbars — analog weight storage + computation
• SRAM-based compute — digital in-memory operations
• Ferroelectric devices — non-volatile, low-power storage + logic

See Also

• Neuromorphics — parent domain
• Memristor — enabling device
• Neuromorphic Hardware — systems using in-memory computing
• Event-Driven Processing — complementary paradigm