Professional Documents
Culture Documents
Yoichi Sato, Kanji Otsuka, Kaoru Kobayashi*, Toshiyuki Kouchi*, Minoru Uwai*, Masato Nishizawa*
Meisei University, Collaborative Research Center
2-1-1 Hodokubo, Hino-shi, Tokyo 191-8506 Japan
Nagase & Co., Ltd
otsuka@gad.meisei-u.ac.jp
ABSTRACT
This is well known that IoT needs intelligent search engines to COMPARISON OF CAM AND AXONERVE
manage big data processing. We have been developing such issues OPERATION STEPS
with novel algorithms.
CAM (Contents Addressable Memory) conducts high speed
Almost all current methods for data searching have been in
searching with high power consumption which is not productive. The
LUT (Look Up Table) algorithm with certain software is one type of data bits coincidence processing. The searching method for
method that is utilized to save power and commonly found in today.. CAM is the typical approach as shown in Figure 2. In one
However there is possibility of a reduced search speed it using the instance, the matched address that is accessed by activation of
sequential software method. The method used in our study produces all of the memory generates the search key data-01101 inputs
high speed searching with low power due to the address-to-data to the search line driver, whereby all the vertical data lines are
search algorithm on LUT when compared with the regular data-to- hit by it at the same time. The address lines are hit by the all
address step. This reversal approach has considered historically. data lines and then a coincident line is activated only.
However address collisions occur prolifically and could not be a Connecting the sense amp drives the right address to encoder.
zero-based probability. In the study of our algorithm developed a Finally the right address (i.e. 01) is produces the outputs for
zero-based probability process which is accomplished by key data the search data of 01101.
division and only division data searching steps resulted in high speed
and low power output.
INTRODUCTION
IoT manages larger and more complex databases. Find data
quickly in a large database can be very difficult and time-
consumption. Smarter search engine will look to find information
quickly with high speed and low power output. One of highest search
engine is CAM (Contents Addressable Memory)[1][2][3] but ti is well
known for its high power consumption due to the need for
simultaneous activity in the entire memory mat as shown in Figure 1
above. Search engine software based nowadays is in the LUT
algorism; however, there is growing dissatisfaction with LUT’s
processing speed. An novel approach developed by our group was
utilized successfully that was named as “AxonerveTM ” with high
speed and low power output as shown in Figure 1 below. That can be FIGURE 2. CAM O PERATION FEATURE IN CASE OF
seen one entry address activations enough for the search. 5BITS DATA
69
Ÿ
70
Ÿ
71
Ÿ
72
Ÿ