In this decade there is huge demand for low power VLSI semiconductor chips. In order to achive low power, power consumption should be minimized at CMOS MOSFET level. In this article, various techniques which are available for minimizing the power consumption at different abstraction levels are discussed in detail. With the help of this article, VLSI design engineers can pick the right methodology with suitable process techniques to reach their required goal.
VLSI Integrated circuits are evolved about four decades ago. VLSI chip designers are still looking for enhancement in the performance of an integrated circuit by accomplishing the less area, low power utilization. scaling techniques has been evolved to achieve these criteria. While scaling the power and area, various challenges encounters those are leakage currents (sub threshold leakage, gate tunneling, hot electron effect, impact ionization, gate induced drain leakage current etc).
Keeping in mind all those performance affecting parameters, VLSI designers produced various design mythologies. There are various techniques to achieve low power design. All this techniques are discussed in detail in this article.
Sources of power consumption in CMOS devices
The total power consumption of CMOS VLSI device can be expressed using equation 1.
Ptotal= Pdynamic + Pshort-circuit + Pstatic…………………………………………………1
A. Dynamic power consumption
Dynamic power consumption is due to charging and discharging of parasitic capacitance.
Energy/transition = CL* VDD2
Power=( Energy/transition)*f*N = CL* VDD2*f*N
Where VDD is power supply, N is the Switching activity; f is the frequency of the signal
B. Short circuit power consumption
Short circuit power consumption is due to the direct current path from VDD to GND while both PMOS and NMOS MOSFET devices are ON for short duration of time. Short circuit power consumption can be mathematically expressed as
Where K is process technology parameter
Vt: threshold voltage
t:rise time (or) fall time
N: average number of transistors in the output stage
f: clock frequency
C. Static power consumption
Static power consumption is due to the leakage current in the MOSFET device.
I1: Reverse bias P-N junction leakage current
I2:Sub-Threshold leakage current
I3 :Gate tunneling current
I4 :Gate induced drain leakage
I5 :Channel punch through current
VLSI design methodology
VLSI Semiconductor IC performance relies upon conflicting parameters such as speed, power consumption, cost and production volume. These contemplations have prodded the improvement in distinct approaches in implementing IC ranging from high performance handcrafted design to fully programmable chip, medium to low performance designs.
Design methodology comparison
1. compact designs.
2.electrical characteristics are improved.
1. Takes lot of time to construct full custom IC.
2. Requires more time for verification
2. Semi custom
1. Standard cells are well tested, which can be directly shared to clients.
2. Best suited for Bottom-up approach.
1. Standard cells requires more time to built. these cells need to be upgraded continuously as technology progress.
2. very expensive in short term but cheaper in long term costs.
Figure below shows VLSI IC design styles
Power reduction techniques at different abstraction levels
There is variety of techniques available to reduce the power consumption of the circuit at different abstraction levels those are mentioned below.
I. At circuit level
There are different VLSI circuits families that exist at this level of abstraction. Those are
Figure below shows example of static CMOS MOSFET logic circuit
In this logic, power consumption can be reduced by using:
2. Ratioed circuits
Pseudo NMOS logic can be used to reduce the power consumption of ratioed circuits. Figure below shows an example of Pseudo NMOS logic.
CVSL logic offers low load capacitance on inputs, no static power consumption and also provides automatic complementary functions.
Dynamic circuits can be implemented using following logics to optimize power.
(a). Domino logic: domino logic adds a buffer at the output of dynamic logic to avoid the affect of output high is pre-charged only with limited drive.
Figure: Generic domino logic
Note: when several logic blocks are cascaded, one clock is not enough to drive entire circuit, solution for this problem is to use two phase clocks as shown below. Use of inverter at the output of dynamic logic can be avoided by using alternate n and p logic blocks.
NP dynamic logic sometime also called as NO Race(NORA)/zipper domino logic.
(b). Dual-rail domino logic: dual rail domino logic is not used for power optimization, because it doubles the number of gates.
(c). Keepers logic: It resembles the same structure as domino, but it improves the performance by introducing keeper transistor at the output of dynamic logic. Keeper transistor will avoid the distribution of pre-charge charge over parasitic capacitance and NMOS array. Figure below shows the example of keeper logic.
(d). Multiple output domino logic (MODL):it enhance the performance by making use of intermediate nodes to form new outputs. But extra circuit may results to slow down the main output. Figure below shows the example of MODL logic.
A. Complementary pass transistor (CPL) is a static gate, because its outputs are connected either to VDD or gnd via low resistance path. This logic facilitates the design of library of gates. Disadvantage is that extra circuit is required to generate differential signals.
b. CMOS with transmission gates: this design enables rail to rail swing. Usually efficient multiplexer designs are implemented using this logic.figure below shows the example of 2:1 mux implemented using this logic.
for more information refer “differential split-level logic for subnanospeed speeds” presented by L.C.M.G Pfennings, IEEE digital library.
Reference: “CMOS Nonthreshold Logic (NTL) and Cascode Nonthreshold Logic (CNTL) for High-speed Applications” by jinn-shyan wang, IEEE digital library
Reference: “Sample set differential logic(SSDC) for complex high-speed VLSI” by T.A Grotjohn, IEEE digital library.
II. At logic level: Many optimization techniques are available at this level, those are:
*Don’t care optimization: circuits at multilevel are optimized by two-level minimization with appropriate don’t care sets. The logic circuit structure may imply that some combination on the inputs of a given logic gate never occur, these combinations are called as satisfiability don’t care sets(SDC) of the gate. Similarly there can be few input combinations for which output valve of the gate is not used in computation of any of the output of the circuit, these set of combinations are known as observability don’t care sets(ODC).These don’t care sets are used for area minimization and switching activity minimization at output of the logic gate.
*path balancing: path balancing is a solution to the problem that aries when there is no equal delay between the inputs of the logic block. When there is no equal delay, output of the logic block will contain spurious tone. Path balancing can be achieved through re-structuring the logic circuit, as depicted in the figure below.
* Logic factorization: A Primary means of technology optimization is the factoring of logic expression. For example consider an expression f=xy+xz+wy+wz this eqation can be factorized into (x+w)(y+z), which results reduced number of transistor count which intern reduces power consumption. Figure below shows the expression is implemented using gates without factorization.
Figure: logic implemented without factorization
Figure: reduced logic after factorization.
III. At gate level: optimization techniques available at this level are
*Technology mapping: technology mapping involves the optimal implementation of a Boolean function using gates from a given library. It serves as the final step in the logic synthesis pipeline. Output of the nodes of the logic circuit mapped to library gates under the cost function of load capacitance multiplied by switching activity. To minimize power dissipation, high switching activity points are either hidden within gates or driven by smaller gates. Minimum power realization under zero delay model can be obtained using dynamic programming.
*Phase assignment : phase assignments inverts the inputs to an operation and, at the same time, also inverting the output. This transformation reduces power in the following ways. First, because this transformation adds inverts on nets that previously did not have inverters, it creates opportunity for several other transformations:two inverters next to each other can be merged and removed, and an inverter at the output of the gate may be absorbed into the gate using a composite gate from the library. Second, it can be used to remove inverters from high-activity nets and move then to low-activity nets.
IV. At RTL level: Optimization techniques available at this level are
*data path re-ordering
* memory partition
V. At architectural level: optimization techniques available at this level are
VI. At system level: optimizing a circuit at this level should undergo following process
*transistor and gate sizing
*super buffer design
*wire sizing, driver sizing and buffer insertion
*clock tree generation