PARALLEL COMPUTING WITH SPIN WAVES - Bio-Inspired and Nanoscale Integrated Computing - page 236

Figure 8.8. Fault recovery example.

the last row) and then from there goes to its destination, as shown in the example
below. We refer to the switches in the extra routing column as switches in column
0. In our example, node 3 connects to row 4 via this switch and reroutes the path
through s(4,6), as shown in Figure 8.8.
The significant advantage of a spin-wave crossbar over a standard VLSI
crossbar is that this simple scheme for rerouting does not collide with other
intercommunications along the same row or column. For instance, as shown in
Figure 8.9, in the same example, node 4 can still send a message to node 5 while 3
sends a message to 6 via s(4,6) in row 4. Although the rerouted path passes
through row 4, these communications can be done in parallel with no conflict.
This is due to the fact that nodes 3 and 4 use different frequencies, so their signal
waves pass through each other without interference.
Note that in this example, the two signals from input 3 and 4 go through row
4, as well as both columns 5 and 6. So the two input messages reach both outputs 5
and 6; however, output 5 detects the message from input 4 on its tuned frequency,
while output 6 receives the signal from input 3.

8.3.2. On Spin-Wave Reconfigurable Mesh
The routing on a reconfigurable mesh is similar to a crossbar. However, this
routing can be from any of the N 2 processing elements to any other one, so there
can exist up to N 2 N 2 different routing schemes.

The routing mechanism in a spin-wave reconfigurable mesh is as follows: To
send information from P i,j to P k,l , the sender, P i,j sends the signal to switch s(i,l) to
be routed to P k,l as shown in Figure 8.10.