&Bullet; physics 14, p72
Many recent advances in computing have been inspired by models of the human brain. For example, researchers have developed a machine learning model that mimics the brain’s ability to recognize new patterns by remembering patterns it has encountered previously. To date, implementations of “associative memory” have mainly included conventional silicon chip-based computers. Now Benjamin Lev of Stanford University and colleagues propose a way to implement associative memory with multiple Bose-Einstein condensates (BECs) and an optical resonator. The researchers say their method of learning and recognizing patterns should be better than the standard associative memory design.
An associative memory computer stores information in a mathematical function that looks like a potential energy landscape with many local minima. Each local minimum corresponds to a separate piece of information. To retrieve this information, the device is initialized in a state close to the relevant minimum and then finds this minimum. This process effectively reconstructs data from inaccurate versions of that data. While everyday technologies typically don’t use associative storage techniques, researchers are interested in them because of their speed and resilience to user error.
The device proposed by the researchers stores information in the energy landscape of several separate BECs located within the same optical cavity. The spin of each BEC interacts with that of the others by scattering photons in the cavity. You can shape the energy landscape of the system by manipulating the position of each BEC. To retrieve information, the BECs are collectively initialized in a particular spin state that relaxes into an energy minimum that is imaged using light emitted from the cavity. The researchers believe that they can build this device in the near future since they have already demonstrated all of the elements of the design.
Sophia Chen is a freelance science writer based in Columbus, Ohio.
- BP Marsh et al., “Improvement of associative memory retrieval and storage capacity through confocal cavity QED”, Phys. Rev. X11, 021048 (2021).