Racetrack memory
Racetrack memory is a proposed non-volatile memory that stores bits as magnetic domains along a nanowire and shifts them past fixed read and write elements with current pulses.
A basic cell has a write head that nucleates a domain in the nanowire, a read head such as a magnetic tunnel junction, and the storage wire, in which domain walls can be pinned by notches. A shift current moves the walls, and with them the data, in either direction. The 2016 review discusses in-plane thermal-stability limits below about 40 nm and perpendicular wall spacing below 90 nm. [2]
The 2008 Science shift-register experiment is a demonstration milestone, not the concept's invention date: Parkin's current-driven magnetic shift-register patent was granted in 2004. [4] [5] Annunziata et al. reported a 256-cell CMOS-integrated array in 2011, with field assistance for the demonstrated read/write operations. This is separate from the later system simulations. [6]
Shift current and wire resistance limit bits per wire at a given supply voltage. Simulated mechanisms give about 128 bits per wire with field assistance (about 10 m/s, 10 pJ/bit), 256 with chiral wall motion (200 m/s, 0.1 pJ/bit) and 64 with voltage-controlled pinning (20 m/s, 1 pJ/bit). [2] In a Co/Pt wire, walls moved at up to 530 m/s at 1 x 10^12 A/m^2 with a threshold of 3.5 x 10^11 A/m^2, and higher densities risk random nucleation and Joule heating. [3]
In a simulated 4-core system, a 64 MB racetrack level-two cache in about the area of 2 MB of SRAM gave 15.8% higher average performance and 73.0% lower cache energy, because static leakage is much lower. One benchmark ran about 7.5% slower because of shift latency, and peripheral circuits took about 20% of area and 40% of latency. [2]
Relations
- related
- NAND flash, Bubble memory, Storage-class memory (SCM), TMR read head, Racetrack position error
Citations
- Perspectives of Racetrack Memory for Large-Capacity On-Chip Memory: From Device to System
- Bi-directional high speed domain wall motion in perpendicular magnetic anisotropy Co/Pt double stack structures
- S. S. P. Parkin: US6834005B1, front page and abstract; granted 21 December 2004
- M. Hayashi et al.: Current-controlled magnetic domain-wall nanowire shift register, Science 320 (2008); original abstract
- A. Annunziata et al.: Racetrack memory cell array with integrated magnetic tunnel junction readout, IEDM 2011; original abstract
Further reading
Sources & caveats
- c2 evaluates device, circuit and cache designs through simulation; its background also cites fabricated research devices. Its system gains are not measurements of a commercial memory.
- c3 is a single-wire laboratory experiment (Scientific Reports, 2017), not a memory device.
- Arithmetic: 8 um / 15 ns = 533 m/s, reported as about 530 m/s.
- c4 is a patent disclosure, not evidence of production hardware or a universal priority date.
- c5 and c6 were checked through IBM's original publication records and abstracts; the 2008 experiment and 2011 CMOS array must not be conflated.
- The reviewed experiments do not establish commercial endurance, retention or cost.