Test Asynchronous FIR Filter Design Presenter PoChun Hsieh

  • Slides: 20
Download presentation
Test Asynchronous FIR Filter Design Presenter: Po-Chun Hsieh Advisor: Tzi-Dar Chiueh Date: 2003/12/1 NTU

Test Asynchronous FIR Filter Design Presenter: Po-Chun Hsieh Advisor: Tzi-Dar Chiueh Date: 2003/12/1 NTU Confidential

Outline • Low Power Issue of Asynchronous Circuits • Test FIR Design - Logic

Outline • Low Power Issue of Asynchronous Circuits • Test FIR Design - Logic circuits - Multiplier - FIR Architecture • Future work • Conclusion • Reference NTU Confidential 2

Low Power Issue of Asynchronous Circuits • No global clock • Functions work only

Low Power Issue of Asynchronous Circuits • No global clock • Functions work only when needed • Dynamic Logic and Domino Logic NTU Confidential 3

Asynchronous FIR filter and other Asynchronous Modules • Every component of FIR works in

Asynchronous FIR filter and other Asynchronous Modules • Every component of FIR works in each evaluation • For other modules, functions work only when needed NTU Confidential 4

Low power Filter • Logic circuits • Multiplier • FIR Architecture NTU Confidential 5

Low power Filter • Logic circuits • Multiplier • FIR Architecture NTU Confidential 5

Computation Units • 4 -phase handshaking is easier to design • Req in is

Computation Units • 4 -phase handshaking is easier to design • Req in is low clean the content “done” is low • Req in is high evaluate “done” is high NTU Confidential 6

Dynamic Logic and Domino Logic • Match the requirements for Computation Unit • No

Dynamic Logic and Domino Logic • Match the requirements for Computation Unit • No spontaneous transitions, low power • Drawbacks – Must be Monotonous Inputs – Completion Detection Methods NTU Confidential 7

Problem of Bounded delay method • How to design delay element ? NTU Confidential

Problem of Bounded delay method • How to design delay element ? NTU Confidential 8

Problem of activity-monitoring completiondetection (AMCD) method (1/2) • Test the transitions at important points

Problem of activity-monitoring completiondetection (AMCD) method (1/2) • Test the transitions at important points • Used in Single-rail CMOS Logic NTU Confidential [1] 9

Problem of activity-monitoring completion-detection (AMCD) method (2/2) • When used in dynamic (domino) logic,

Problem of activity-monitoring completion-detection (AMCD) method (2/2) • When used in dynamic (domino) logic, maybe it will never pull down the signal NTU Confidential [2] 10

Differential Cascode Voltage Switch Logic (DCVSL) • Dual-rail Domino Logic gate • Drawbacks: Area、Power

Differential Cascode Voltage Switch Logic (DCVSL) • Dual-rail Domino Logic gate • Drawbacks: Area、Power consumption • Completion detection method only add in the output cascode stages NTU Confidential 11

Low power Logic gate • Single-rail bounded-delay dynamic (domino) logic gates are very low

Low power Logic gate • Single-rail bounded-delay dynamic (domino) logic gates are very low power, but have two problems • Before finding solutions, choose to use DCVSL NTU Confidential 12

Asynchronous Multiplier • Multiplier is the most important part in a FIR filter •

Asynchronous Multiplier • Multiplier is the most important part in a FIR filter • Sign and Magnitude [5] NTU Confidential 13

Data Dependent Carry Save Adder Array (1/2) • MD: Multiplicand; MR: Multiplier; PP: Partial

Data Dependent Carry Save Adder Array (1/2) • MD: Multiplicand; MR: Multiplier; PP: Partial Product NTU Confidential [3] 14

Data Dependent Carry Save Adder Array (2/2) • 4 X 4 Carry Save Adder

Data Dependent Carry Save Adder Array (2/2) • 4 X 4 Carry Save Adder array • By probability, only turn on 50% adders NTU Confidential [4] 15

Partially work by DCVSL Logic • If MR(n)=“ 1” ( MR(n). t=“ 1”; MR(n).

Partially work by DCVSL Logic • If MR(n)=“ 1” ( MR(n). t=“ 1”; MR(n). f=“ 0”; ) then the adder cell works • If MR(n)=“ 0” ( MR(n). t=“ 0”; MR(n). f=“ 1”; ) then the adder cell will not work NTU Confidential 16

FIR Architecture • H: Handshake Circuit; h(0)~h(N): coefficients NTU Confidential 17

FIR Architecture • H: Handshake Circuit; h(0)~h(N): coefficients NTU Confidential 17

Future Work • Test the designed FIR module • Search for low power FIR

Future Work • Test the designed FIR module • Search for low power FIR architecture and Multiplier • Try to find solutions to the problems of single-rail dynamic (domino) logic NTU Confidential 18

Conclusion • Unlike other kinds of Asynchronous circuits, every component of FIR works every

Conclusion • Unlike other kinds of Asynchronous circuits, every component of FIR works every time. • Choose FIR architecture, Multiplier, and Implement Circuits to make FIR low power. NTU Confidential 19

Reference [1] Grass, E. and S. Jones, "Activity-monitoring completion- detection (AMCD): a new approach

Reference [1] Grass, E. and S. Jones, "Activity-monitoring completion- detection (AMCD): a new approach to achieve self-timing", Electronics Letters, vol. 32, no. 2, pp. 86 -88, January 1996 [2] Bartlett, V. A. and E. Grass, "Completion-detection technique for dynamic logic, " Electronics Letters, vol. 33, no. 22, pp. 1850 -1852, October 1997. [3] Bartlett, V. A. and E. Grass, “A Self-Timed Multiplier using Condutional Evaluation", Proc. PATMOS'98, 8 th International Workshop on Power, Timing, Modelling, Optimization and Simulation, Lyngby, Denmark, pp. 429 -438, October 1998 [4] D. Kearney and N. W. Bergmnn, “Bundled Data A syncheonous Multipliers with Data Dependent Computation Times”, Proc. ASYNC’ 97, 2 nd Int. Symp. On Advanced Research in Asynchronous Circuits and Systems, pp. 186 -197, 1997 [5] Bartlett, V. A. and E. Grass, “A Low-Power Asynchronous VLSI FIR Filter", Proc. ARVLSI'01, 19 th Conference on Advanced Research in VLSI, Salt Lake City, Utah, USA, March 2001. NTU Confidential 20