Information Technology Reference
In-Depth Information
7.6.2. Optical Routing
The optical routing in ORM is the routing through optical free-space interconnec-
tions. The data transfer does not use any electrical bus in the system. All N 2
processors can communicate in unit time delay as long as there is only one read or
write from or to each location. In the following, we describe how such an optical
connection is established between two processors through the RM.
A connection phase consists of two cycles. In the first cycle, each processor
sends the address of its desired destination processor to the arithmetic control unit
of its associated mirror using its dedicated laser. The arithmetic control unit of the
mirror computes a rotation degree such that both the source and the destination
processors have equal angle with the line perpendicular to the surface of the
mirror in the plane formed by the mirror, the source processor, and the destination
processor. Once the angle is computed, the mirror is rotated to point to the desired
destination. In the second cycle, the connection is established by the laser beam,
carrying the data from the source to the mirror and then from the reflected mirror
towards the destination. The read operation has two phases. In the first phase, the
read requirement and the reader's address are sent to the processor which stores
the desired data. In the second phase, the data is sent back to the reader,
depending on the reader's address. Both phases use the two-cycle write routing
method.
7.6.3. Electro-Optical Routing
This communication mechanism establishes an efficient full connectivity among
only the N processors situated diagonally in the processing layer on the N 2
processors in the ORM (i.e., for processors P(j, j) where 1
N). This routing
technique uses electrical buses on the processing layer and fixed mirrors on the
deflection layer.
The connection for electro-optical routing is implemented as follows. Each
processor P(j, j) is associated with the jth row of the deflection unit, where the row
contains N fixed mirrors. The ith fixed mirror in that row for 1
j
r
r
N is directed
to the processing unit P(i, i). In an exclusive read exclusive write (EREW) scenario,
any processing element P(i, i) sends data to P(k, k) in the following way: First,
P(i, i) sends the data to P(i, k) through the electrical row bus; then, P(i, k) sends
data to P(k, k) through the transmitter and its deflector M(i, k). The variety of
techniques available in this architecture makes ORM a very powerful computing
model [15].
In our proposed hierarchical multiscale architectures, the processing elements
are interconnected using spin waves at the nanoscale level, and via ORM's
electrical, optical, or electro-optical routing at the microlevel [16]. Figure 7.12
shows this organization.
As an example, an electro-optical routing is shown in Figure 7.12. In this
routing, as mentioned previously no reconfiguration of mirrors is necessary, and
only fixed mirrors are used. The data communication in each row is through
i
r
r
 
Search WWH ::




Custom Search