1. June 6, 20071 Using Negative Edge Triggered FFs to Reduce Glitching Power in FPGA Circuits Tomasz S. Czajkowski and Stephen D. Brown Department of Electrical and Computer Engineering University of Toronto, Ontario, Canada

2. 2 Motivation  Glitches: Undesirable logic transitions that occur due to delay imbalance in the logic circuit Waste power and do not provide any useful functionality Can increase the average toggle rate of a net by as much as a factor of 2 Not well defined until post placement and routing  Glitches can be filtered out by strategically inserting negative edge triggered FFs

3. 3 Glitches in FPGAs  Due to unequal arrival time of signals at the inputs of LUTs  Glitches can be propagated through LUTs 4LUT Generated Propagated

4. 4 Reducing Glitches  Insert a negative edge triggered FF after a LUT that produces or propagates glitches 4LUT Generated clock No glitches

5. 5 Alternatives  Gated D-latch Implement a gated D-latch in a LUT Input signal is transparent during the latter half of the clock period  Gated LUT Gate the output of a LUT with the clock input using an AND or an OR gate Similar effect as gated D-latch Can generate glitches too  When implemented Gated D-latch consumes 50% more power than a FF and double that of a gated LUT Neither alternative is very effective

6. 6 Background on Dynamic Power  Average Net Dynamic Power Dissipation P avg is average power V is supply voltage f clock is the clock frequency s i is the average per cycle toggle rate of a net C i is the capacitance of a net

7. 7 Power Model  Goal To be able to compute the change in dynamic power dissipation in the logic elements affected by a negative edge triggered FF insertion  Power dissipated by a LUT and a FF  Toggle Rate of logic signals (s i )  Net capacitance (C i )

8. 8 LUT Power  The LUT itself dissipates an non- trivial amount of power when its inputs toggle  We look at how the power dissipated by a LUT relates to the frequency of its output transitions

9. 9 LUT Power Model

10. 10 FF Power  How much power would it cost to insert a FF into a circuit?  What about the power cost of alternatives to a FFs? Gated LUT Gated D-latch

11. 11 Clocked Element Power Comparison

12. 12 Toggle Rate of Logic Signals  Topic is covered considerably in literature  Toggle rate model based on the concept of Transition Density [Najm’94] and the work of Anderson and Najm [AN’03] The latter work decomposes transition density into transitions generated by a LUT and that propagated through a LUT.  Modified to include delay information in order to account for glitches

13. 13 Examples of Wires P[y]P t (y) P[y’=1 | y=0] P[y’=0 | y=1] D(y) D(y) – P t (y) ½11110 ½½≈0.4 ½0 1/8¼ 1¼0 ¼ 1½¼ Clock A B C D

14. 14 Wire Properties NameDescriptionNotation Static Probability Probability that a wire assumes the logic value 1 in any given clock cycle. P[y] Transition Probability The average number of state transitions, excluding glitches. P t (y) Low to High Transition Probability Probability that a wire will change state to logic value 1, given that it is at a logic value 0 at present. P[y’=1 | y=0] High to Low Transition Probability Probability that a wire will change state to logic value 0, given that it is at a logic value 1 at present. P[y’=0 | y=1] Transition Density The average number of logic value transitions per cycle. Includes glitches. D(y) Average Number of Glitches per cycle The average number of useless transitions per clock cycle D(y)-P t (y)

15. 15 Propagating Glitches Through a LUT  Increase D(z) to account for glitches that occur on wire y (D(y)-P t (y)). Do so only when x remains at constant 1 for the duration of the clock cycle. y x z

16. 16 Estimate Error

17. 17 Net Capacitance  We need to be able to estimate net capacitance to figure out the difference in dynamic power dissipation due to a change in the transition density of a net  Relate net capacitance (unavailable directly) to net delay (available through timing report) Distinguish between nets of different fanout

18. 18 Fanout 1 Net Capacitance

19. 19 Fanout 2 Net Capacitance

20. 20 Fanout 3 Net Capacitance

21. 21 Fanout 4 Net Capacitance

22. 22 Higher Fanout Net Capacitance  In our benchmark set fewer than 5% of the nets had fanout greater than 4 Clock net is excluded from calculation  Approximate capacitance of net with fanout n>4 as:  Not exact, but supports the fact that glitches on nets with high fanout are bad Average estimate error of +22%

23. 23 Negative Edge Triggered FF Insertion Algorithm 1. Scan all nets in a logic circuit to determine if negative edge FF insertion can be applied 2. Analyze the resulting set of nets to determine the benefit of applying the optimization to each net (determined by the cost function) 3. Apply the optimization to a net on which the most power could be saved 4. Repeat until no beneficial choices are found

24. 24  Compute change in power (∆P) + cost of adding a FF - power saved on the modified net - power saved on nets and LUTs in the transitive fanout of the added FF  Compute the change in the minimum clock period (∆T) Specify ∆T allowed (∆T a )  where u(x) is the step function  Accept change when ∆C < 0 Cost Function

25. 25 Example LUT Some logic network LUT FF

26. 26 Example: Inserted FF LUT Some logic network LUT FF Neg FF

27. 27 Example: Compute change in the # of glitches LUT Some logic network LUT FF Neg FF

28. 28 Example: Compute change in the # of glitches LUT Some logic network LUT FF Neg FF

29. 29 Example: Compute change in LUT power dissipation LUT Some logic network LUT FF Neg FF

30. 30 Experimental Results  8 benchmark circuits taken from QUIP package  Synthesize, place, route and analyze timing of a circuit using Quartus II 5.1  Apply algorithm to reduce glitches in a circuit Aim to decrease the minimum clock period by no more than 5%  Perform timing analysis once the circuit has been modified  Use ModelSIM-Altera 6.0c for simulation Simulate a circuit both pre- and post- modification using the same clock frequency  Use PowerPlay Power analyzer to estimate the average dynamic power dissipation of each circuit

31. 31 Experimental Results Circuit name Simulation Clock Frequency (MHz) Minimum Clock PeriodDynamic Power Dissipation Initial (ns) Final (ns) Change (%) Initial (mW) Final (mW) Change (%) Barrel64*2004.3864.8068.74229.94189.7-17.50 mux64_16bit2753.052 0389.24 0.00 fip_cordic_rca1257.5517.8513.8243.2839.49-8.76 oc_des_perf_opt2902.9893.072.641058.8796.7-24.75 oc_video_compression_ systems_huffman_enc 2603.626 094.8895.190.33 cf_fir_24_8_81705.3755.715.87290.41292.90.84 aes128_fast1406.2516.5694.84879.24870.6-0.99 rsacypher1406.3766.5632.8550.7348.22-4.95 Average +3.6-7.0

32. 32 Observations (1)  oc_des_perf_opt Large number of XOR gates present Removing glitches from one node removes a lot of glitches on the nodes in its transitive fanout (up to the next FF)  mux64_16bit The cost function determined that no net was a good candidate for optimization Very few glitches were present in the circuit and the power they dissipate was not large enough to warrant the insertion of FFs

33. 33 Observations (2)  cf_fir_24_8_8 Overestimated toggle rate caused the algorithm to apply negative edge triggered FF insertion too excessively Need to include spatial correlation in the toggle rate model  aes128_fast Toggle rate is 50% higher than in oc_des_perf_opt Most nets use local LAB connections, causing little power dissipation Insertion of 173 FFs only achieved 1% power reduction  Saved 35.14 mW in routing alone, because toggle rate on all affected wires was reduced by 50-70%  Added 24.6 mW due to FF insertion  Added 1.86 mW to the power dissipated by the clock network, because new LABs were connected to the clock network  Net win of 8.68 mW

34. 34 Conclusion  Negative edge triggered FF insertion can work well to reduce glitches in a circuit Computing glitches propagated to the transitive fanout of a net is important, especially when XOR gates are present When inserting a lot of negative edge triggered FFs, be mindful where they go. Do target LABs have a clock signal already routed to them?  Unlike retiming, our approach only needs to ensure that exactly one negative edge triggered FF is on any given combinational path Retiming may require the translation of more than a single FF to be valid

35. 35 Future Work  Better toggle rate prediction algorithm that includes spatial correlation  Having FFs that can be negative edge triggered without using an additional LAB clock line would make the cost of this optimization lower Silicon area cost vs. frequency of use trade-off

36. 36 Acknowledgement  We’d like to express our gratitude to Altera for funding this research  We’d like to thank Altera Toronto in particular for dedicating some of their time to answer our questions and provide insight throughout the course of this work

37. June 6, 200737 Questions?