Spin-Transfer Torque

Spin-transfer torque is the physical mechanism that lets a current flowing through a magnetic tunnel junction flip the free layer's magnetization directly, with no external magnetic field involved at all — a spin-polarized electron stream transfers its own angular momentum to the free layer's electron spins as it passes through, and above a critical current density that transferred torque overcomes the layer's own magnetic stability and forces it to reorient.

Current direction, not just current magnitude, determines which state the free layer ends up in, which is the detail that makes STT switching fundamentally different from thermally-assisted or field-driven writes. Reversing the current's direction reverses which layer's electrons do the polarizing and which layer receives the torque, so a current flowing one way pushes the free layer toward parallel alignment with the fixed layer, and current flowing the other way pushes it toward antiparallel — this bidirectional, current-direction-encoded write is why STT-MRAM needs only a single bidirectional current source per cell rather than the separate word-line field-generating circuitry earlier field-switched MRAM designs required.

Critical current density scales down favorably with cell size, which is the specific physical property that made STT viable at modern transistor dimensions where field-switched MRAM had already run out of room. The torque transferred per electron doesn't change with device size, but a smaller magnetic tunnel junction has proportionally less total magnetic volume to switch, so the absolute current needed to reach critical density drops as the cell shrinks — field-switched MRAM instead needed current through nearby wires scaled to generate a sufficient field strength independent of the target cell's own size, which hit a wall as neighboring cells packed close enough that one cell's field-switching current started disturbing its neighbors. STT's switching current scaling downward with the device, rather than staying roughly fixed, is the single physical fact that let MRAM keep scaling past the point field-switching could.

Switching speed and switching current trade against each other, and that tradeoff is where most real STT-MRAM circuit design effort goes. Driving more current through the junction switches the free layer faster, but higher write current also means higher power consumption and, past a point, risks degrading the thin tunnel barrier through repeated electrical stress; designers balance this by choosing a target write time — often tens of nanoseconds for general-purpose embedded memory — and sizing the write current to just clear the critical density with margin at that speed, rather than maximizing switching speed outright, since doing so would demand current levels the tunnel barrier's long-term reliability can't sustain.

Read spin-transfer torque through a *the-electron-stream-is-the-write-mechanism* lens rather than a *just-another-MRAM-detail* lens: no external field, no separate write wire, no field-focusing structure — a single bidirectional current through the junction itself carries both the energy and the directional information needed to set the bit, which is precisely the simplification that let MRAM's cell design keep shrinking after field-switching ran out of room.