Stanford vector matrix multiplier

Damman grating to replicate the source inputs. The use of the CGH to replicate the inputs comes from the CGH property that the spots in the replay field are the Fourier transform of input illumination. Since only one channel is likely to be required at each output, those not required can be blocked using liquid crystal shutters. Such switches are based on the Stanford vector matrix multiplier (SVMM) [50] related switching devices [51]. When implemented using a CGH to fan out, and with a 2D array of inputs (rather than the 1D arrays of the SVMM) to simplify the free space optics, these are called matrix - matrix switches [52]. This kind of structure is found in a range of optical processing architectures (see Sec. 2.4). For a symmetrical switch with n inputs and n outputs, an array of nxn shutters is required. 

The most significant development, which led to the use of liquid crystal devices in neural networks, was the Stanford vector matrix multiplier (SVMM) by Goodman in 1978 [50, 63]. The basic structure of the system is shown in Fig. 56. 

---