Thermal Aware Energy Efficient Content Addressable Memory Design on FPGA

Thermal Aware Energy Efficient Content Addressable Memory Design on FPGA

TarandeepKaur, AmanpreetKaur, BishwajeetPandey

Chitkara University Research and Innovation Network (CURIN) Chitkara University, Punjab Campus

Chandigarh, India

AbstractIn this work, we are going to integrate the thermalaware and energy efficient design approach in content addressable memories. During implementation on FPGA, we are going to test the thermal stability by operating this CAM design with different ambient temperature 250C, 350C, 450C, 550C, and 650C. We have shown the compatibility of our device with wireless network by operating it on different frequency ranges: 1GHz, 2.4 GHz, 3.6 GHz, 4.9 GHz, 5 GHz, 5.9 GHz, 60 GHz

bands of WLAN channels. There is 69.38%, reduction in leakage power, when we scale down ambient temperature from 650C to 550 C, 450 C, 350 C, 250 C at 1 GHz. In this work, we are using 28nm technology based Kintex-7 FPGA and Verilog Hardware Description Language.

KeywordsThermal Aware, Energy Efficient, FPGA, Content Addressable Memory.

1. INTRODUCTION

   Content Addressable Memory (CAM) is an application specific memory which allows us to search the content in the single clock cycle [1]. Content addressable memory is also known by the name associative memory, associative array [7] or associative storage [2]. Unlike the other computer memories that uses memory address for the reference of data word [6] stored at that particular address, Content addressable memory (CAM) is a special type of computer memory used in a very high speed searching applications in which user supply the data word and CAM searches the complete memory in order to find the data word in the memory.

   Fig. 1: Top Level Schematic of CAM

   In Figure1, this CAM has three inputs and two outputs. If data word is found successfully, then CAM returns the complete list of addresses (Storage addresses) where the data word was

   found. Content Addressable Memory (CAM) was mainly designed to search the entire memory in single operation due to which it is much faster than the other memories (mainly RAM) used [2]. Kintex-7 is 28nm technology based FPGA.

   Fig. 2: 28nm Technology Based Kintex-7 FPGA

   Kintex-7 FPGA has maximum number of logic cells in compare to other 90nm technology based Virtex-4 FPGA, 65nm technology based Virtex-5 FPGA and 45nm technology based Spartan-6 FPGA. Kintex-7 FPGA has Block RAM, DSP Slices, PCI Gen2 Blocks, GTX Transceivers, I/O Pins as shown in Table 1. Random Access Memory (RAM) uses simple storage cells, On other hand CAM have its individual comparison circuit in order to identify a match between the i/p(input) bit and the stored bit. Also matched o/p (output) from individual cell in the data word should be combined to produce a complete data word match signal. Due to this extra circuitry area of Content addressable memory chip increases making manufacturing cost high .The parallel nature of CAM for searching in which huge amount of circuitry is active on every cycle make it more power consumptive [11].

   TABLE 1: DEVICE RESOURCE AVAILABLE IN KINTEX- 7 FPGA

   Because of these disadvantages, CAM is only used in some special applications where the slow speed is not tolerable. Two types of CAM mainly used are Binary CAM and Ternary CAM [1]. Binary CAM simply uses data search words that contains 1s and 0s to perform the exact match, But in Ternary CAM there exists a third state known as ' Don't Care' or 'X' which make our searching more flexible. For e.g. Ternary CAM can have a stored word "X010X" which will match any of these words "10101", "10100", "00101", "00100". It makes our came more flexible but also increases the cost of it. Content addressable memory frequently in computer networking device (CND)[3]. For example, when a switchinghub (also known as network switch) receives data frame (DF) from any of the ports , internal table is updated with the DF's source MAC address

   [4] and port it was received on, it then looks for the destination's MAC address in the table to determine what port the frame needs to be forwarded to, and send it out on that port. MAC address table is mainly implemented using binary CAM (BCAM) so destination port can be found very quickly Ternary Cam (TCAM) are mainly used in routers [5].

   Fig. 3: Resource Utilization (%) of CAM on FPGA

   When we are implementing CAM on Kintex-7 FPGA, then CAM is using 1% of slice LUTs, 1% of Slice Registers, 89% of IO and 3% of clocking available on Kintex-7 FPGA. With the use of TCAM in routing table, lookup process becomes very efficient [8]. The use of "don't care" to store the addresses for the host part of address make it easy and quick to look up the destination address in CAM for immediate retrieval of the correct routing entry; both comparison and masking are done by the Content addressable memory hardware [2]. Thermal aware is the demand of time, so we are operating CAM on different

   ambient temperatures: 250 C, 350C, 450C, 550C, and 650C.

   The design we are representing is compatible with Wireless network channels using IEEE 802.11 protocols at different frequency range from 2.4GHz to 60 GHz. W are using 298.15K in place of 250C and similarlywe are

   increasing degree Celsius by 273.15 and convert into Kelvin for thermal analysis.

   TABLE 2: LIST OF WLAN 802.11 CHANNELS

   WLAN Channels	Frequency (GHz)	Range and Specification
   802.11b/g/n	2.4 GHz	2412-2484MHz
   802.11y	3.6 GHz	3657.5-3692.5MHz
   802.11a/h/j/n/ac	5 GHz	4915-5825MHz
   802.11p	5.9 GHz	5850-5925MHz
   802.11ad	60 GHz	WiGig

2. THERMAL ANALYSIS OF CAM

   1. When CAM is Operating at 1 GHz Frequency

      TABLE 3: POWER DISSIPATION AT DIFFERENT TEMPERATURE

      Temperature Power	250 C	350 C	450 C	550 C	650 C
      Clock	0.008	0.008	0.008	0.008	0.008
      Logic	0.005	0.005	0.005	0.005	0.005
      Signal	0.024	0.024	0.024	0.024	0.024
      IO	0.028	0.028	0.028	0.028	0.028
      Leakage	0.045	0.057	0.076	0.104	0.147
      Total	0.11	0.122	0.141	0.169	0.211

      There is 69.38%, reduction in leakage power, when we scale down ambient temperature from 650C to 550 C, 450 C, 350 C, 250 C at 1 GHz as shown in Figure 4 and

      Table 3.

      Fig. 4: Power Dissipation at 1GHz Frequency and Different Temperature

   2. When CAM is Operating at 2.4 GHz Frequency

      TABLE 4: POWER DISSIPATION AT DIFFERENT TEMPERATURE

      Temperature Power	250 C	350 C	450 C	550 C		650 C
      Clock	0.019	0.019	0.019	0.019	0.019
      Logic	0.011	0.011	0.011	0.011	0.011
      Signal	0.058	0.058	0.058	0.058	0.058
      IO	0.069	0.069	0.069	0.069	0.069
      Leakage	0.046	0.058	0.076	0.105	0.148
      Total	0.203	0.215	0.234	0.262	0.305

      There is 68.91%, reduction in leakage power, when we scale down ambient temperature from 650C to 550 C, 450 C, 350 C, 250 C at 2.4 GHz as shown in Figure 5 and

      Table 4.

      Fig. 5: Power Dissipation at 2.4GHz Frequency and Different Temperature

   3. When CAM is Operating at 3.6 GHz Frequency

      TABLE 5: POWER DISSIPATION AT DIFFERENT TEMPERATURE

      Temperature Power	250 C	350 C	450 C	550 C	650 C
      Clock	0.029	0.029	0.029	0.029	0.029
      Logic	0.017	0.017	0.017	0.017	0.017
      Signal	0.087	0.087	0.087	0.087	0.087
      IO	0.104	0.104	0.104	0.104	0.104
      Leakage	0.046	0.058	0.077	0.106	0.148
      Total	0.283	0.295	0.314	0.343	0.385

      There is 68.91%, reduction in leakage power, when we scale down ambient temperature from 650C to 550 C, 450 C, 350 C, 250 C at 3.6 GHz as shown in Figure 6 and Table 5.

      Fig, 6: Power Dissipation at 3.6GHz Frequency and Different Temperature

   4. When CAM is Operating at 4.9 GHz Frequency

      TABLE 6: POWER DISSIPATION AT DIFFERENT TEMPERATURE

      Temperature Power	250 C	350 C	450 C	550 C	650 C
      Clock	0.039	0.039	0.039	0.039	0.039
      Logic	0.023	0.023	0.023	0.023	0.023
      Signal	0.119	0.119	0.119	0.119	0.119
      IO	0.142	0.142	0.142	0.142	0.142
      Leakage	0.046	0.058	0.077	0.106	0.149
      Total	0.369	0.381	0.400	0.429	0.473

      There is 69.12%, reduction in leakage power, when we scale down ambient temperature from 650C to 550 C, 450 C, 350 C, 250 C at 4.9 GHz as shown in Figure 7 and Table 6.

      Fig. 7: Power Dissipation at 4.9GHz Frequency and Different Temperature

   5. When CAM is Operating at 5 GHz Frequency.

      TABLE 7: POWER DISSIPATION AT DIFFERENT TEMPERATURE

      Temperature Power	250 C	350 C	450 C	550 C	650 C
      Clock	0.040	0.040	0.040	0.040	0.040
      Logic	0.023	0.023	0.023	0.023	0.023
      Signal	0.121	0.121	0.121	0.121	0.121
      IO	0.145	0.145	0.145	0.145	0.145
      Leakage	0.046	0.058	0.077	0.106	0.149
      Total	0.376	0.388	0.407	0.436	0.479

      There is 69.12% reduction in leakage power, when we scale down ambient temperature from 650C to 550 C, 450 C, 350 C, 250 C at 5 GHz as shown in Figure 8 and Table

      7.

      Fig. 8: Power Dissipation at 5GHz Frequency and Different Temperature

   6. When CAM is Operating at 5.9 GHz Frequency

      TABLE 8: POWER DISSIPATION AT DIFFERENT TEMPERATURE

      Temperature Power	250 C	350 C	450 C	550 C	650 C
      Clock	0.047	0.047	0.047	0.047	0.047
      Logic	0.028	0.028	0.028	0.028	0.028
      Signal	0.143	0.143	0.143	0.143	0.143
      IO	0.172	0.172	0.172	0.172	0.172
      Leakage	0.046	0.058	0.078	0.107	0.150
      Total	0.436	0.448	0.467	0.496	0.540

      There is 69.33%, reduction in leakage power, when we scale down ambient temperature from 650C to 550 C, 450 C, 350 C, 250 C at 5.9 GHz as shown in Figure 9 and

      Table 8.

      Fig. 9: Power dissipation at 5.9GHz Frequency and Different Temperature

   7. When CAM is operating at 60 GHz Frequency

   TABLE 9: POWER DISSIPATION AT DIFFERENT TEMPERATURE

   Temperature Power	250 C	350 C	450 C	550 C	650 C
   Clock	0.482	0.482	0.482	0.482	0.482
   Logic	0.111	0.111	0.111	0.111	0.111
   Signal	1.325	1.325	1.325	1.325	1.325
   IO	1.777	1.777	1.777	1.777	1.777
   Leakage	0.054	0.070	0.096	0.134	0.190
   Total	3.749	3.765	3.791	3.829	3.885

   There is 71.57%, reduction in leakage power, when we scale down ambient temperature from 650C to 550 C, 45 0 C, 350 C, 250 C at 60 GHz as shown in Figure 10 and Table 9.

   Fig. 10: Power Dissipation at 60GHz Frequency and Different Temperature

3. CONCLUSION

   We have achieved thermal aware energy efficient Content addressable memory on FPGA. We operated our device on different ambient temperatures and check for its working with Wireless network channels using IEEE 802.1 protocols at different frequency range from 1 GHz to 60 GHz. We have achieved an energy efficient CAM as we are getting 71-68% reduction in leakage power, when we scale

   down ambient temperature from 650C to 550 C, 450 C, 350 C, 250 C. Leakage power is directly proportional to

   temperature. Although, there is no reduction in Clock power, Logic power, Signal power, IO power but there is reduction in Total power.

4. FUTURE SCOPE

In future we will use content addressable memory with the concept of pervasive computing. We have used 2D IC in this paper but we will implement the design in future using 3D, 4D IC. We have implemented this CAM using 28 nm FPGA, but there is a large scope of implementing CAM on 14nm, 20nm, 32nm, 40nm, 90nm and large scale FPGA. This design is energy efficient but we can make it High Performance also. This design is implemented using Verilog. In future, we can also go for VHDL, System C, Embedded C, and C++ for high level synthesis.

REFERENCES

1. I. Arsovski, T. Chandler, A. Sheikholeslami, A ternary content- addressable memory (TCAM) based on 4T static storage and including a

   current-race sensing scheme, IEEE Journal of Solid-State Circuits, 38(1), 155-158.

2. Content Addressable memory. http://en.wikipedia.org/wiki/Content- addressable_memory. Accessed on September 2014.

3. Computer Networking Device, Accessed on September 2014 http://en.wikipedia.org/wiki/Networking_hardware.

4. MAC address, http://en.wikipedia.org/wiki/MAC_address. Accessed on September 2014

5. Router, http://en.wikipedia.org/wiki/Router_(computing). Accessed on September 2014

6. Data Word,

   http://en.wikipedia.org/wiki/Word_(computer_architecture). Accessed on September 2014

7. Associative Array, http://en.wikipedia.org/wiki/Associative_array.

   Accessed on September 2014

8. K. Pagiamtzis, & A. Sheikholeslami, Content-addressable memory (CAM) circuits and architectures: A tutorial and survey, Solid-State Circuits, IEEE Journal of, 41(3), 712-727, 2006.

9. R. D. Adams, Content Addressable Memories. High Performance Memory Testing: Design Principles, Fault Modeling and Self-Test, pp. 67-75.

10. H. Miyatake, M. Tanaka, & Y. Mori, A design for high-speed low- power CMOS fully parallel content-addressable memory macros, Solid-State Circuits, IEEE Journal of, 36(6), 956-968, 2001.

11. K. Pagiamtzis, &A.Sheikholeslami, A low-power content-addressable memory (CAM) using pipelined hierarchical search scheme, IEEE Journal of Solid-State Circuits, 39(9), 1512-1519, 2004.