Hardware Reference
In-Depth Information
Table 3.17 Performance of MX-1
Process technology
90 nm 7Cu CMOS low standby process
Core size
3.1 mm 2
Operation frequency
200 MHz@1.2v
Power consumption
250 mW
Maximum performance
40GOPS
(16b addition)
3.6GOPS
(Fixed point signed 16bit MAC)
H-Ch0
H-Ch1
4b ALU
V-Ch
-8x
-4x
0
4x
8x
Mux
-2x
-x
0
x
2x
Adder
XREG
[3:0]
Booth
Encoder
0
[3:2]
Booth
Encoder
1
[1:0]
Fig. 3.70
Circuit diagram of 4-bit-grained PE
3.3.2
MX-2
The required performance of image processing is getting higher and higher; there-
fore, the second version of MX core (MX-2), which architecture is improved from the
MX-1, is developed [ 63 ]. The main technologies for the enhancement are men-
tioned below:
1. Expanding the processor elements from 2-bit grained to 4-bit grained
2. Improving the pipeline architecture of the MX controller
3. Equipping double frequency mode
Hereafter, these technologies are described in detail.
Figure 3.70 shows a block diagram of the PE of the MX-2. The PE contains a
4-bit temporary register (XREG) and 4-bit-grained ALU. The XREG loads data from
data registers through Horizontal channel (H-ch0, H-ch1) or Vertical channel (V-ch).
The PE loads data from data registers and through H-ch0 or H-ch1 and operates
Search WWH ::




Custom Search