Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Low Power VLSI Designs

1,836 views

Published on

Low Power design for VLSI is becoming paramount important since its the most contributing factor to the performance analysis.Therefore power aware VLSI designs, Physical aware power reduction techniques and low power VLSI architectures are vital for future

Published in: Engineering
  • Great projects with clear instructions. I've gotten an enormous feeling of accomplishment pride from making my own bookshelves and side tables. Thanks Ted! โ—โ—โ— https://url.cn/ktFCrsHZ
       Reply 
    Are you sure you want to  Yes  No
    Your message goes here
  • There are over 16,000 woodworking plans that comes with step-by-step instructions and detailed photos, Click here to take a look โ– โ– โ–  http://t.cn/A6hKwqcb
       Reply 
    Are you sure you want to  Yes  No
    Your message goes here
  • Want to preview some of our plans? You can get 50 Woodworking Plans and a 440-Page "The Art of Woodworking" Book... Absolutely FREE โ™ฃโ™ฃโ™ฃ http://tinyurl.com/yy9yh8fu
       Reply 
    Are you sure you want to  Yes  No
    Your message goes here
  • Thank you!!!!very helpful
       Reply 
    Are you sure you want to  Yes  No
    Your message goes here

Low Power VLSI Designs

  1. 1. Intelligent Systems and Advanced Computing (iSaac) 15 LOW POWER VLSI DESIGN VLSI Power Architecture Mahesh Dananjaya
  2. 2. VLSI POWER IN A NUTSHELL VLSI POWER ARCHITECTURE W A T Mahesh Dananjaya Department of Electronic and Telecommunication Engineering University of Moratuwa Sri Lanka
  3. 3. VLSI POWER IN A NUTSHEL Content 1. Semiconductor Engineering 1.1. Semiconductors 1.1.1.Extrinsic Semiconductors 1.2. Transistors 1.2.1.Bipolar Junction Transistors (BJT) 1.2.2.Field Effect Transistors (FET) 1.2.2.1. Junction Field Effect Transistors (JFET) 1.2.2.2. Metal Oxide Semiconductor Field Effect Transistors (MOSFET) 1.2.2.2.1. MOS Capacitance 1.2.2.2.2. Basic MOSFET Operation 1.2.2.2.3. Enhancement Mode & Depletion Mode 1.2.2.2.4. P Channel MOSFET 1.2.2.2.5. N Channel MOSFET 1.3. Basic Logic Cells 2. Power Estimation 2.1. Dynamic Power 2.1.1.Switching Power 2.1.2.Internal Power 2.2. Static Power 2.2.1.Leakage Power 2.2.1.1. Threshold Leakage 2.2.1.2. Sub-threshold Conduction 2.2.1.3. Transistor Leakage Mechanisms 2.2.1.3.1. Drain Leakage Current/Sub threshold Leakage Current 2.2.1.3.1 Reverse Biased Current 2.2.1.3.1 Sub Threshold Drain Current 2.2.1.3.1 Gate Induced Drain Leakage 2.2.1.3.2. Gate Leakage Current 2.2.1.3.1 Gate Tunneling 2.2.1.3.2.1.1. Fowler-Nordheim Tunneling 2.2.1.3.2.1.2. Direct Tunneling 2.2.1.3.1 Hot Carrier Injection 2.3. Technology Scaling & Power Estimation 2.3.1.CMOS Device Scaling 2.3.1.1. Gate Oxide Thickness Scaling 2.3.1.2. Channel Miniaturization 2.3.1.3. Supply Voltage and Threshold Voltage Scaling 2.3.1.4. Doping Concentration
  4. 4. 2.3.1.5. Source Drain Punchthrough 2.4. Miscellaneous Power Consumption and Estimation 2.4.1.Meta-stability 2.4.1.1. Hold Time 2.4.1.2. Set Up Time 2.4.2.Glitches 2.4.3.Latch-ups 2.5. Electronic Design Automation (EDA) Power Estimation Techniques 3. Power Reduction 3.1. Dynamic Power Reduction 3.1.1.Clock Gating 3.1.1.1. Local Clock gating 3.1.1.1.1. Latch Based Clock gating 3.1.1.1.2. Latch Free Clock Gating 3.1.1.2. Multi-Level Boolean Logic 3.1.1.2.1 Satisfiability Donโ€™t Care (SDC) 3.1.1.2.1 Observability Donโ€™t Care (ODC) 3.1.1.3. Advanced Clock Gating 3.1.1.3.1 Stability Condition (STC) 3.1.1.3.1 Observability Donโ€™t Care (ODC) 3.1.1.4. Enable Strengthening and Clock Gating Efficiency 3.1.1.4.1. Strong STC 3.1.1.4.2. Strong ODC 3.1.1.5. Memory Power Reduction 3.1.1.5.1. Redundant Read Removal 3.1.1.5.2. Redundant Write Removal 3.1.1.5.3. Memory As Steering Point for Register Power Reduction 3.1.1.5.4. Light Sleep Power Reduction 3.2. Static Power Reduction 3.2.1.Comparison with Dynamic Power 3.2.2.Power Gating 3.2.2.1. Sleep Transistors/Switches 3.2.2.1.1. Switch Types 3.2.2.1.2. Switch Sizing 3.2.2.1.3. Switch Placing Architecture 3.2.2.2. Power Gating Modes 3.2.2.2.1. Fine Grain Power Gating 3.2.2.2.2. Coarse Grain Power Gating 3.2.2.3. Power Gating Factors 3.2.2.4. Synchronous and Asynchronous Power Gating 3.2.2.5. Power Gating/Controlling Mechanisms
  5. 5. 3.2.2.5.1. Non-State Preserving Power Gating 3.2.2.5.2. State Preserving Power Gating 3.2.2.5.2.1.1. State retention Techniques 3.2.2.5.2.1.1.1. Retention Registers 3.2.2.6. Compiler Based and Hardware Based Power Gating 4. Power Verification 5. Power Fix
  6. 6. Introduction to VLSI Power Electronics Electronic has been emerging its counterpart ever since the semiconductor was born. Now it has been developed throughout few decades and changed world into upside down. Electronic has evolved from vacuumed tubes to highly complex electronics designs passing through many more intermediate technology revolutions such as diodes, transistors, microcontrollers, microprocessors, ASIC, DSP, FPGA, SOC, SOPC, MPSOCs. Integrated Circuit designs have dramatically changed to very low level designs to Very Large Scale Integrated Circuit (VLSI) designs. These technology systems consist of internal technology nodes such as logic cells, Gates, adders, Flip-flops, counters, registers, multiplexers etc. Electronic systems became the caliber behind every nook and corner of the technology and provide solutions to every field such as computing, robotics, biomedical engineering, IOT, embedded system designing ,Telecommunication and information processing, networking and security etc. Nowadays technologies serve in a system level and advancing the inherent strength of the electronic fields to optimize and improve the system performances. Electronic Systems Electronic system is set of interacting and independent electronic components including active as well as passive components. System design can be identified in different levels. Component/Device level consists of electronic components and semiconductor components such as transistors and diodes. In medium level abstraction these designs can be identified as circuit level such as logic level. In high level abstraction electronic designs can be identified as a System level where Main system is consist of such sub systems which integrated together. Therefore the overall system performances are depending on the optimizations of each and every sub system and integrating them in the optimized way. VLSI Power VLSI power is the main area that all the future electronic designs and developments looking forward to, because of its essence to core performances and values. System power estimation and reduction is the major focus of most of the design engineers today. Therefore throughout this paper I am trying to give overall perception and a core impression about system power. Because all the electronic designs are focused on novel power reducing techniques which can save IC power by a large percentage. Clock gating, Power gating are some of basic techniques used to reduce integrated circuit power. Electronic engineers are working on all these layers to reduce power such as device level, circuit level and high level abstraction levels.
  7. 7. VLSI Power Estimation, Reduction, Fix and Verification 1. Semiconductor Engineering 1.1 Semiconductors Material that has both conduction and insulation properties are called a semiconductor. As the name implies the semiconductors are materials of having conductivity between insulators and pure conductors. Operation of semiconductors is based on the quantum physics and energy band theories. Materials such as Silicon, Germanium are well known semiconductors. 1.1.1 Extrinsic Semiconductors Base Semiconductor is doped with P-type impurities such as Boron (B), Aluminum (Al) and N-type impurity materials such as Phosphorus (P) and Arsenic (As). 1.2 Transistors
  8. 8. Transistors are the semiconductor devices which can be used as an electronic switches, amplifiers and power sources. There are two types of transistors, Bipolar Junction Transistors (BJT) and Field Effect Transistors (FET). 1.2.1 Bipolar Junction Transistors (BJT) Bipolar Junction Transistors are relies on the contact and combining of two types of semiconductor for its operation. BJT can be used as amplifier, switches and oscillators. Operation of BJT involves both electrons and holes. Uniquely in semiconductors these two type s are originated with the two different doping of impurities in to base material. Current flow is due to diffusion of charge carriers across the junction between two regions of different charge concentrations. Electrons are majority charge carriers in n-type semiconductors, whereas holes are majority charge carriers in p-type semiconductors. Basically BJTs are consisting of 3 lead called Base, Collector and Emitter. Therefore we can use characteristics equations and graphs to identify the operation of the BJT. By controlling the electric field across Base-Emitter junction thus depletion layer we can control the charged carriers flows between collector and emitter. Above figure shows the operation of NPN transistor. Characteristics graph of the BJTs are shown here. This will illustrate the operation of NPN transistor. PNP transistor also works in a similar way.
  9. 9. Transistors are used in three different configurations, common Base, common Collector and common collector. We are not discussed them here and our aim is to illustrate power of those semiconductor materials later. Unipolar Transistors 1.2.2 FET (Field Effect Transistor) Field Effect Transistor (FET) is one of unipolar transistors that use an electric field to control the shape and hence the conductivity of a channel of one type of charge carrier in a semiconductor material and involve single carrier operation. FET are single charged carrier device and FET is consist of an active channel through which charge carriers, electrons and holes, flow from the source to drain. Source and drain terminal conducts are connected to the semiconductor through Ohmic contacts. More important characteristic of this device is that the conductivity of the channel is a function of the potential applied across the gate and source terminals. There are three leads for FET, gate, source and drain. Field Effect Transistor (FET) operates as a conducting semiconductor channel with two Ohmic contacts between semiconductor and metal, the source and the drain, where the amount of charged carriers the channel is controlled by the third contact, the gate. In the vertical direction the gate channel and the substrate structure (gate junction) can be regarded as an orthogonal two terminal device, which is either a MOS structure or a reverse biased rectifying device that controls the mobile charge across the channel by capacitive coupling (Filed Effect). Metal Oxide Semiconductor FET (MOSFET), junction gate FET (JFET), Metal Semiconductor (MESFET), Heterostructure FET (HFET), are types of FETs. 1.2.2.1 JFET or JUGFET (Junction Gate Field Effect Transistors)
  10. 10. JFET is a type of FET and simplest type of field effect transistors. JFETs are used as electronically controlled switches, amplifiers or voltage controlled resistors. According to the channel formed by the source and drain there are two types. N type and P type. This JFET has large input impedance. The JFET is a long channel of semiconductor material, doped to contain an abundance of positive charged carriers or holes (p-type), or of negative carriers or electrons (n-type).Ohmic contact at each end form the source (S) and drain (D).A PN Junctions formed on one or both sides of the channel, or surrounding it, using a region with doping opposite to that of the channel, and biased using an ohmic gate contact (G). JFET is known as a depletion mode device. Flow of electric charge through a JFET is controlled by constricting the current-carrying channel. The current also depends on the electric field between source and drain (analogous to the difference in pressure on either end of the hose). Constriction of the conducting channel is accomplished using the field effect. Voltage between the gate and source is applied to reverse bias the gate-source PN-junction. Therefore drain to source current is controlled by the electric field between gate to source. Characteristic diagram of the JFET is shown below.
  11. 11. 1.2.2.2 MOSFET (Metal Oxide Semiconductor) The most important FET is the MOSFET. In a silicon MOSFET, the gate contact is separated from the channel by an insulating silicon dioxide layer. The charge carrier of the conducting channel constitute an inversion charge, that is, electrons in the case of a P-type substrate in n channel device and holes in the case of a n-type substrate in P channel devices, induced in the semiconductor at the silicon insulator interface by the voltage applied across the gate electrodes. The electrons enter and exit the channel at n+ source and drain contacts regarding the N channel MOSFET and at p+ regarding the P channel MOSFET. 1.2.2.2.1 MOS Capacitance MOSFET is different from the JFET operation because rather than having a reverse biased rectifying features MOS structure has a MOS capacitor which make a bridge between gate and the
  12. 12. substrate layer. MOS capacitance is the vital part of MOSFET which constitutes the important gate- channel-substrate structure of the MOSFET. MOS capacitor is a two terminal semiconductor device of practical interest in its own right. As shown in the below figure it consist of a metal contact separate from the semiconductor by a dielectric insulator. An additional Ohmic contact is provided at the semiconductor substrate. Most of the time MOS structure use doped silicon as the substrate and its native oxide, silicon oxide as the insulator part. Because of the insulator it creates an infinite input impedance, preventing any charged carrier transport across the dielectric layer when a bias voltage is applied between the metal and the semiconductor. Instead, the applied voltage will induce charge and counter charges and counter charges in the metal and in the interface layer of the semiconductor, similar to what we expect in the metal plates of a conventional parallel plate capacitor. However in the MOS capacitor we may use the applied voltage to control the type of interface charge we induce in the semiconductor, majority carriers, minority carriers and depletion charge. 1.2.2.2.2 Basic MOSFET Operation In the MOSFET an inversion layer at the semiconductor oxide interface acts as a conducting channel. For example in an n channel MOSFET, the substrate is a p-type silicon and the inversion charge consist of electrons that form a conducting channel between the n+ ohmic source and the drain contacts. At DC condition, the depletion region and the neutral substrate provide isolation between devices and fabricated on the same substrate.
  13. 13. In MOS capacitor, inversion charge can be induced in the channel by applying a suitable gate voltage relative to other terminals. The onset of strong inversion is defined in terms of a threshold voltage Vt being applied to the gate electrode relative to the other terminals. In order to assure that the induced inversion channel extends all the way from source to drain, it is essential that the MOSFET gate structure either overlaps slightly or align with the edges of these contacts. Self-alignment is preferable since it minimize the parasitic gate-source and gate-drain capacitances. 1.2.2.2.3 MOSFET- Depletion Mode and Enhancement Mode So far we have discussed about the paramount importance of the MOSFET to the VLSI technology and without MOSFETs contribution modern electronic systems consist of VLSI technology and large scale integration is impossible. MOSFETs are usually capable of voltage gain and signal power gain. In this section we will try to get some understanding about the MOSFET modes and operations which are very deeply contribute to the VLSI power architectural analysis. N channel and P channel MOSFETs are the two basic types of MOSFETs. In n channel MOSFET current is due to the flow of electrons in inversion layer and in p channel current is due to the flow of holes. There are two modes of MOSFET based on their design architecture, Depletion mode and Enhancement mode. Due to various reasons such as performance, power and efficiency enhancement mode is widely used in VLSI designs.
  14. 14. 1.2.2.2.4 P Channel MOSFET MOSFET which deploy p channel region between source and drain is called p-channel MOSFET. Usually itโ€™s a four terminal device gate, source, drain and substrate or body. Mostly substrate lead is connected with source lead. Drain and the source are heavily doped with p type impurities thus called p+ regions and substrate is a doped n-type region. Current flow due to flow of positively charged holes thatโ€™s why itโ€™s called p channel MOSFET. Depletion Mode: There is a p-type region beneath the oxide layer thus already has a current flowing path and already form a junction between n-type substrate materials. Therefore itโ€™s really effects the characteristic MOSFET. Enhancement Mode: There is no p-type region beneath the insulation part and thus threshold voltage is higher than the depletion mode type. Weak inversion layer is established by the holes extracted by the p+ regions such as source and drain and minority carriers of the n type substrate. MOSFET Enhancement Mode N Channel P Channel Depletion Mode N Channel P Channel
  15. 15. When applying negative gate voltage, the electrons present beneath the oxide layer, experiences repulsive force and they are pushed downwards into the substrate, the depletion region is populated by the bound positive charges which are associated with the donor atoms. Provided that negative voltage to the gate attracts holes from p+ source and drain region in to the channel region. Thus the channel is formed with the holes. Now if we apply a voltage difference between the source and the drain, current is flowing through the channel. The gate voltage controls the Hole concentration of the channel. Only different in the enhancement mode and depletion mode is that depletion mode MOSFET already contain a conduction channel (path) and has inversion layer, but enhancement mode do not have a inversion layer and thus form a weak inversion layer when negative voltage is supplied to the gate by extracting p (holes) from p+ regions. Thus compose a conduction channel and when gate to source voltage exceeds the threshold, conduction happens from source to drain if we provide a drain to source voltage. Characteristic curves of P channel MOSFET P channel Enhancement Mode Operation
  16. 16. P Channel Depletion Mode 1.2.2.2.5 N Channel MOSFET MOSFET having an n-channel region between source and drain is known as n-channel MOSFET. N channel MOSFET is also consist with four terminals, source, drain, gate and substrate or body. The drain and source are heavily doped n+ and the substrate is p-type. The current flows due to flow of the negatively charged electrons. Thus it is called an n-channel MOSFET. Depletion Mode: There is a n-type region beneath the oxide layer thus already has a current flowing path and already form a junction between p-type substrate materials. Therefore itโ€™s really effects the characteristic MOSFET.
  17. 17. Enhancement Mode: There is no n-type region or channel beneath the insulation part and thus threshold voltage is higher than the depletion mode type. There is no inversion layer and the weak inversion layer is established by the holes extracted by the n+ regions such as source and drain and minority carriers of the p type substrate.
  18. 18. Comparison Curves NMOS, PMOS, CMOS MOS logic is built upon the MOSFET and especially CMOS which is widely used in VLSI designs. NMOS is N type MOSFET and PMOS is a P type MOSFET. CMOS is the combination of p-type and n-type which is widely used in VLSI power designs. CMOS is the combining PMOS and NMOS MOSFETs to design MOS logic. CMOS Inverter is shown below. Basic Logic Cell JFET MOSFET-Depletion MOSFET-Enhancement
  19. 19. CMOS Inverter Threshold Voltage Sub Threshold Leakage and Sub-threshold Current
  20. 20. 2. Power Estimation Integrated circuit design has evolved through some technology clicks and nowadays designers are capable of VLSI (Very Large Scale ICs) with billions of transistors. Therefore there may be hundreds of ways for power consuming processes inside the integrated circuits. But we can organize them into three major categories and some few exceptional cases. Switching power, internal power and leakage power are the major ways of power consumption. These three can be classified into two Types according to their nature, Dynamic power and Static power. Dynamic power consists of switching and internal power components and Static power is compromised with a Leakage power. Power hierarchy is illustrated below. 2.1 Dynamic Power Dynamic power is dissipated due to toggling effects or state change of the signals. Due to assertion and desertion of the signal on the signal output capacitance of the output pin will charge and discharge, thus generating a dynamic power. There are two major components that contribute for the dynamic power. They are Switching power and Internal Power. System Power Dynamic Power Switching Power Internal Power Static Power Leakage Power
  21. 21. ๐ท๐‘ฆ๐‘›๐‘Ž๐‘š๐‘–๐‘ ๐‘ƒ๐‘œ๐‘ค๐‘’๐‘Ÿ = ๐‘†๐‘ค๐‘–๐‘ก๐‘โ„Ž๐‘–๐‘›๐‘” ๐‘ƒ๐‘œ๐‘ค๐‘’๐‘Ÿ + ๐ผ๐‘›๐‘ก๐‘’๐‘Ÿ๐‘›๐‘Ž๐‘™ ๐‘ƒ๐‘œ๐‘ค๐‘’๐‘Ÿ ๐‘ƒ ๐ท = ๐‘ƒ๐‘† + ๐‘ƒ๐ผ ๐‘คโ„Ž๐‘’๐‘Ÿ๐‘’ , ๐‘ƒ ๐ท = ๐ท๐‘ฆ๐‘›๐‘Ž๐‘š๐‘–๐‘ ๐‘ƒ๐‘œ๐‘ค๐‘’๐‘Ÿ ๐‘ƒ ๐ท = ๐ท๐‘ฆ๐‘›๐‘Ž๐‘š๐‘–๐‘ ๐‘ƒ๐‘œ๐‘ค๐‘’๐‘Ÿ ๐‘ƒ๐‘† = ๐‘†๐‘ค๐‘–๐‘ก๐‘โ„Ž๐‘–๐‘›๐‘” ๐‘ƒ๐‘œ๐‘ค๐‘’๐‘Ÿ ๐‘ƒ๐ผ = ๐ผ๐‘›๐‘ก๐‘’๐‘Ÿ๐‘›๐‘Ž๐‘™ ๐‘ƒ๐‘œ๐‘ค๐‘’๐‘Ÿ 2.1.1 Switching Power Switching power is generated because of the state changes and toggling of the output signal, thus charging and discharging the capacitance of the output net. Switching power is based on the load capacitance. Main factors which effects the switching power intensity id illustrated below, o Clock frequency o Activity switching frequency Clock frequency is that frequency that your system clock run and it may be injected into every flop of the design. Therefore clock is very important parameter to calculate switching power. Activity switching frequency is the other important parameter and it is depends on our operations and may change to instance by instance. Activity data is very important parameter to modern EDA tool which will be capable of power estimation and reductions. Mostly these activity data is extracted from the simulation files.
  22. 22. Switching power calculation is based on the following principle and factors. o C -Size of the total load capacitance Higher capacitance means more charge (more power) it requires to pull the line to correct voltage. o ๐‘ฝ โˆ’Driving Voltage Higher voltage means we need more and more charge (Q) o ๐’‡ โˆ’ ๐’๐ฐ๐ข๐ญ๐œ๐ก๐ข๐ง๐  ๐…๐ซ๐ž๐ช๐ฎ๐ž๐ง๐œ๐ฒ When the switching frequency is high more charge-discharge cycles are happening. Thus more power is released. ๐‘ท ๐‘บ = ๐‘ช โˆ— ๐‘ฝ ๐Ÿ โˆ— ๐’‡ 2.1.2 Internal Power Power dissipated the gate when the inputs are charging is an internal power. Internal power is based on the input capacitance and internal power is also belongs to the dynamic power category. This is the short circuit stage of power and ground at the transition stage, where both NMOS and PMOS conducting for a short period of time. Thus generating a short circuit current. As a brief internal power is the consuming power due to temporary short circuit paths and internal capacitance Factors that affect the internal power is, o Input edge time o Slew Rate o Internal Capacitance o Load Capacitance
  23. 23. Calculation of the internal power depends on the two factors. o ๐‘‰ โ€“ ๐ท๐‘Ÿ๐‘–๐‘ฃ๐‘–๐‘›๐‘” ๐‘‰๐‘œ๐‘™๐‘ก๐‘Ž๐‘”๐‘’ o ๐ผ๐‘†๐ถ โˆ’ ๐‘†โ„Ž๐‘œ๐‘Ÿ๐‘ก ๐ถ๐‘–๐‘Ÿ๐‘๐‘ข๐‘–๐‘ก ๐ถ๐‘ข๐‘Ÿ๐‘Ÿ๐‘’๐‘›๐‘ก ๐‘ท ๐‘ฐ = ๐‘ฝ โˆ— ๐‘ฐ ๐‘บ๐‘ช Therefore Dynamic power can be calculated as the sum of switching power and internal power. And we can optimized the result ๐‘ท ๐‘ซ = ๐‘ท ๐‘บ + ๐‘ท ๐‘ฐ ๐‘ท ๐‘ซ = ๐‘ช โˆ— ๐‘ฝ ๐Ÿ โˆ— ๐’‡ + ๐‘ท ๐‘ฐ ๐‘ท ๐‘ซ = ๐‘ช โˆ— ๐‘ฝ ๐Ÿ โˆ— ๐’‡ + ๐‘ฝ โˆ— ๐‘ฐ ๐‘บ๐‘ช ๐‘ท ๐‘ซ โฉญ ๐‘ช ๐’†๐’‡๐’‡ โˆ— ๐‘ฝ ๐Ÿ โˆ— ๐’‡ ๐’”๐’˜๐’Š๐’•๐’„๐’‰
  24. 24. 2.2 Static Power Static power is dissipated due to non-ideal characteristic of the transistor and totally the leakage power. But there is a true potential of reducing unwanted and unnecessary power consumptions via other power reduction techniques based on the static power. We consider Static Power dissipated when the device is sitting ideal and itโ€™s totally a leakage power. At this state the semiconductor devices are powered on but switched off. Therefore it will leads to a leakage of the devices. Although devices are expected to be on ideal, according to the quantum physics and Firmy-Dirac equation there may be leakage through the channel of the semiconductor. Static power is considerably lower than the dynamic power. But due to some important reasons static power is becoming more and more important part of the VLSI power estimation. Static power is there because of the non-ideal characteristic of the device where it act as a large resistor allowing a small current to flow through the device. ๐‘†๐‘ก๐‘Ž๐‘ก๐‘–๐‘ ๐‘ƒ๐‘œ๐‘ค๐‘’๐‘Ÿ = ๐ฟ๐‘’๐‘Ž๐‘˜๐‘Ž๐‘”๐‘’ ๐‘ƒ๐‘œ๐‘ค๐‘’๐‘Ÿ ๐‘ท ๐‘บ = ๐‘ฝ โˆ— ๐‘ฐ๐’๐’†๐’‚๐’Œ 2.2.1 Leakage Power Power dissipated when the semiconductor device is powered on. This is also known as a sub- threshold leakage most of the time. But leakage is not only consisting with sub-threshold leakage, but also another components. Some factors affect the leakage power consumption is illustrated below. 2.2.1.1 Threshold Leakage and Threshold Voltage ๐‘ฝ๐’•๐’‰ โˆ’ Threshold Voltage Threshold Voltage of Field Effect Transistor (FET) is the minimum gate to source voltage (๐‘‰๐‘”๐‘ ) difference that is needed to create a conducting path between source and drain terminals. This concept
  25. 25. is widely used in MOSFET in enhancement mode and depletion mode. We already analyzed that depletion mode MOSFET have lower ๐‘ฝ ๐’•๐’‰ than enhancement mode MOSFET. At gate to source voltage above he threshold voltage (๐‘‰๐‘”๐‘  > ๐‘‰๐‘กโ„Ž), but still below saturation, the transistor is in linear region. This stage also known as Ohmic region where it act like a voltage controlled variable resistor. When reffering to the junction field transistor (JFET), the threshold voltage is often called โ€œPinch- Off Voltageโ€ instead. 2.2.1.2 Sub threshold Conduction Subthreshold Conduction, subthreshold leakage or subthreshold drain current is the current between the source and drain of a MOSFET when the transistor is in subthreshold region or weak inversion region. Thus the gate to source voltage below the threshold voltage. Most of the time this subthreshold conduction is considered as the parasitic leakage in a state ythat would ideally have no current. On the other hand weak inversion region is an efficient operating region and sub threshold is a useful transistor mode around which circuit functions are designed. But with the rapid technology revolutions leakage power consumption and subthreshold leakage are becoming vital to VLSI architecture. 2.2.1.3 Transistor Leakage Mechanisms With the rapid increase in the transistor density, exponential growth of the number of on chip active devices, the continuous decrease of the ๐‘‰๐‘กโ„Ž โˆ’ ๐‘กโ„Ž๐‘Ÿ๐‘’๐‘ โ„Ž๐‘œ๐‘™๐‘‘ ๐‘ฃ๐‘œ๐‘™๐‘ก๐‘Ž๐‘”๐‘’ and reduction of the ๐‘‡๐‘œ๐‘ฅ gate oxide thickness result in a significant amount of leakage power consumption in a VLSI design architecture in deep sub-micron technologies. Totally static power is consumption is generated by leakage current which flows when the device is not ideal. Leakage power is becoming a critical and vital concern for the designers particularly for circuits that have low-duty cycles, bursty operation and rely on batteries for long period of time. Therefore potential of having a leakage in MOS devices is paramount important to VLSI power architecture. Leakage current is not just a sub-threshold leakage, but consist of several parasitic components. Basically the leakage can be categorized into two categories. ๏‚ท ๐ผ ๐‘‚๐น๐น โˆ’ ๐ท๐‘Ÿ๐‘Ž๐‘–๐‘› ๐ฟ๐‘’๐‘Ž๐‘˜๐‘Ž๐‘”๐‘’ ๐ถ๐‘ข๐‘Ÿ๐‘Ÿ๐‘’๐‘›๐‘ก โ€“ ๐‘†๐‘ข๐‘ โˆ’ ๐‘กโ„Ž๐‘Ÿ๐‘’๐‘ โ„Ž๐‘œ๐‘™๐‘‘ ๐ฟ๐‘’๐‘Ž๐‘˜๐‘Ž๐‘”๐‘’ ๐ถ๐‘ข๐‘Ÿ๐‘Ÿ๐‘’๐‘›๐‘ก (๐‘“๐‘™๐‘œ๐‘ค๐‘  ๐‘“๐‘Ÿ๐‘œ๐‘š ๐‘‘๐‘Ÿ๐‘Ž๐‘–๐‘› ๐‘ก๐‘œ ๐‘ ๐‘œ๐‘ข๐‘Ÿ๐‘๐‘’ ๐‘œ๐‘Ÿ ๐‘๐‘œ๐‘‘๐‘ฆ) o ๐ผ ๐ท,๐‘ค๐‘’๐‘Ž๐‘˜ โˆ’ ๐‘†๐‘ข๐‘ โˆ’ ๐‘กโ„Ž๐‘Ÿ๐‘’๐‘ โ„Ž๐‘œ๐‘™๐‘‘ ๐ท๐‘Ÿ๐‘Ž๐‘–๐‘› ๐ถ๐‘ข๐‘Ÿ๐‘Ÿ๐‘’๐‘›๐‘ก o ๐ผ๐‘–๐‘›๐‘ฃ โˆ’ ๐‘…๐‘’๐‘ฃ๐‘’๐‘Ÿ๐‘ ๐‘’ ๐ต๐‘–๐‘Ž๐‘ ๐‘’๐‘‘ ๐ถ๐‘ข๐‘Ÿ๐‘Ÿ๐‘’๐‘›๐‘ก o ๐ผ ๐บ๐ผ๐ท๐ฟ โˆ’ ๐บ๐‘Ž๐‘ก๐‘’ ๐ผ๐‘›๐‘‘๐‘ข๐‘๐‘’๐‘‘ ๐ท๐‘Ÿ๐‘Ž๐‘–๐‘› ๐ฟ๐‘’๐‘Ž๐‘˜๐‘Ž๐‘”๐‘’ ๏‚ท ๐ผ ๐บ๐ด๐‘‡๐ธ โˆ’ ๐บ๐‘Ž๐‘ก๐‘’ ๐ฟ๐‘’๐‘Ž๐‘˜๐‘Ž๐‘”๐‘’ ๐ถ๐‘ข๐‘Ÿ๐‘Ÿ๐‘’๐‘›๐‘ก
  26. 26. (๐ฟ๐‘’๐‘Ž๐‘˜๐‘Ž๐‘”๐‘’ ๐‘๐‘ข๐‘Ÿ๐‘Ÿ๐‘’๐‘›๐‘ก ๐‘กโ„Ž๐‘Ž๐‘ก ๐‘‘๐‘Ÿ๐‘–๐‘๐‘๐‘™๐‘’ ๐‘กโ„Ž๐‘Ÿ๐‘œ๐‘ข๐‘”โ„Ž ๐‘กโ„Ž๐‘’ ๐‘”๐‘Ž๐‘ก๐‘’๐‘  ๐‘œ๐‘“ ๐‘กโ„Ž๐‘’ ๐‘ก๐‘Ÿ๐‘Ž๐‘›๐‘ ๐‘–๐‘ ๐‘ก๐‘œ๐‘Ÿ) o ๐ผ ๐‘‡๐‘ˆ๐‘๐‘๐ธ๐ฟ โˆ’ ๐บ๐‘Ž๐‘ก๐‘’ ๐‘‡๐‘ข๐‘›๐‘›๐‘’๐‘™๐‘–๐‘›๐‘” o ๐ผ ๐ป๐ถ โˆ’ ๐ป๐‘œ๐‘ก ๐ถ๐‘Ž๐‘Ÿ๐‘Ÿ๐‘–๐‘’๐‘Ÿ ๐ผ๐‘›๐‘—๐‘’๐‘๐‘ก๐‘–๐‘œ๐‘› 2.2.1.3.1 Sub-threshold Leakage Current (๐‘ฐ ๐‘ถ๐‘ญ๐‘ญ) The sub-threshold leakage current of a transistor, ๐ผ ๐‘‚๐น๐น, is defined as the drain current when |๐‘‰๐‘”| โˆ’ |๐‘‰๐‘ | = 0 and ๐‘‰๐ท โ‰ฅ 0. ๐ผ ๐‘‚๐น๐น is dependent on the various factors such as, ๏‚ท ๐‘‰๐ท๐ท โˆ’ ๐‘†๐‘ข๐‘๐‘๐‘™๐‘ฆ ๐‘‰๐‘œ๐‘™๐‘ก๐‘Ž๐‘”๐‘’ ๏‚ท ๐‘‰๐‘กโ„Ž โˆ’ ๐‘‡โ„Ž๐‘Ÿ๐‘’๐‘ โ„Ž๐‘œ๐‘™๐‘‘ ๐‘‰๐‘œ๐‘™๐‘ก๐‘Ž๐‘”๐‘’ ๏‚ท Doping Concentration ๏‚ท ๐‘‡๐‘‚๐‘‹ โˆ’ ๐บ๐‘Ž๐‘ก๐‘’ ๐‘‚๐‘ฅ๐‘–๐‘‘๐‘’ ๐‘‡โ„Ž๐‘–๐‘›๐‘’๐‘ ๐‘  As we mentioned earlier ๐‘ฐ ๐‘ถ๐‘ญ๐‘ญ is consist of 3 sub leakage components. o ๐ผ ๐ท,๐‘ค๐‘’๐‘Ž๐‘˜ โˆ’ ๐‘†๐‘ข๐‘ โˆ’ ๐‘กโ„Ž๐‘Ÿ๐‘’๐‘ โ„Ž๐‘œ๐‘™๐‘‘ ๐ท๐‘Ÿ๐‘Ž๐‘–๐‘› ๐ถ๐‘ข๐‘Ÿ๐‘Ÿ๐‘’๐‘›๐‘ก o ๐ผ๐‘–๐‘›๐‘ฃ โˆ’ ๐‘…๐‘’๐‘ฃ๐‘’๐‘Ÿ๐‘ ๐‘’ ๐ต๐‘–๐‘Ž๐‘ ๐‘’๐‘‘ ๐ถ๐‘ข๐‘Ÿ๐‘Ÿ๐‘’๐‘›๐‘ก o ๐ผ ๐บ๐ผ๐ท๐ฟ โˆ’ ๐บ๐‘Ž๐‘ก๐‘’ ๐ผ๐‘›๐‘‘๐‘ข๐‘๐‘’๐‘‘ ๐ท๐‘Ÿ๐‘Ž๐‘–๐‘› ๐ฟ๐‘’๐‘Ž๐‘˜๐‘Ž๐‘”๐‘’ ๐‘ฐ ๐‘ถ๐‘ญ๐‘ญ = ๐ผ๐‘–๐‘›๐‘ฃ + ๐ผ ๐ท,๐‘ค๐‘’๐‘Ž๐‘˜ + ๐ผ ๐บ๐ผ๐ท๐ฟ
  27. 27. 2.2.1.3.1.1 ๐‘ฐ๐’Š๐’๐’— โˆ’ ๐‘น๐’†๐’—๐’†๐’“๐’”๐’† ๐‘ฉ๐’Š๐’‚๐’”๐’†๐’… ๐‘ช๐’–๐’“๐’“๐’†๐’๐’• ๐‘ฐ๐’Š๐’๐’— is the current that flows through the reverse biased diode between the drain and the p region of the transistor, and it is dependent on the junction area between the Source/Drain terminal and the body and exponentially dependent to the temperature. Leakage current for the inverse biased diode can be modelled as follows where, o ๐‘ˆ ๐‘‡ โˆ’ ๐‘‡โ„Ž๐‘’๐‘Ÿ๐‘š๐‘Ž๐‘™ ๐‘‰๐‘œ๐‘™๐‘ก๐‘Ž๐‘”๐‘’ (๐‘ƒ๐‘Ž๐‘Ÿ๐‘Ž๐‘š๐‘’๐‘ก๐‘ค๐‘Ÿ ๐‘™๐‘–๐‘›๐‘’๐‘Ž๐‘Ÿ๐‘™๐‘ฆ ๐‘‘๐‘’๐‘๐‘’๐‘›๐‘‘๐‘’๐‘›๐‘ก ๐‘œ๐‘› ๐‘กโ„Ž๐‘’ ๐‘ฃ๐‘œ๐‘™๐‘ก๐‘Ž๐‘”๐‘’) o ๐ผ๐‘† โˆ’ ๐‘…๐‘’๐‘ฃ๐‘’๐‘Ÿ๐‘ ๐‘’ ๐‘†๐‘Ž๐‘ก๐‘ข๐‘Ÿ๐‘Ž๐‘ก๐‘–๐‘œ๐‘› ๐ถ๐‘ข๐‘Ÿ๐‘Ÿ๐‘’๐‘›๐‘ก (๐ผ๐‘›๐‘ก๐‘Ÿ๐‘–๐‘›๐‘ ๐‘–๐‘ ๐‘ƒ๐‘Ž๐‘Ÿ๐‘Ž๐‘š๐‘’๐‘ก๐‘’๐‘Ÿ ๐‘“๐‘œ๐‘Ÿ ๐‘กโ„Ž๐‘’ ๐‘‘๐‘’๐‘ฃ๐‘–๐‘๐‘’) o ๐‘‰๐ท โˆ’ ๐‘‰๐‘œ๐‘™๐‘ก๐‘Ž๐‘”๐‘’ ๐‘๐‘’๐‘ก๐‘ค๐‘’๐‘’๐‘› ๐ท๐‘Ÿ๐‘Ž๐‘–๐‘› ๐‘Ž๐‘›๐‘‘ ๐‘กโ„Ž๐‘’ ๐ต๐‘œ๐‘‘๐‘ฆ๐‘œ๐‘“ ๐‘กโ„Ž๐‘’ ๐‘ก๐‘Ÿ๐‘Ž๐‘›๐‘ ๐‘–๐‘ ๐‘ก๐‘œ๐‘Ÿ o ๐ฝ๐ผ๐‘๐‘‰ โˆ’ ๐‘†๐‘Ž๐‘ก๐‘ข๐‘Ÿ๐‘Ž๐‘ก๐‘œ๐‘–๐‘› ๐ถ๐‘ข๐‘Ÿ๐‘Ÿ๐‘’๐‘›๐‘ก ๐ท๐‘’๐‘›๐‘ ๐‘–๐‘ก๐‘ฆ o ๐ด ๐ท โˆ’ ๐ท๐‘–๐‘“๐‘“๐‘ข๐‘ ๐‘–๐‘œ๐‘› ๐ด๐‘Ÿ๐‘’๐‘Ž Then, ๐‘ฐ ๐‘ฐ๐‘ต๐‘ฝ = ๐‘ฐ ๐‘บ(๐’†(๐‘ฝ ๐‘ซ ๐‘ผ ๐‘ปโ„ ) โˆ’ ๐Ÿ) ๐‘ฐ ๐‘ฐ๐‘ต๐‘ฝ = ๐‘จ ๐‘ซ โˆ— ๐‘ฑ ๐‘ฐ๐‘ต๐‘ฝ 2.2.1.3.1.2 ๐‘ฐ ๐‘ซ,๐’˜๐’†๐’‚๐’Œ โˆ’ Sub-Threshold Drain Current When,๐‘‰๐‘” < ๐‘‰๐‘กโ„Ž, |๐‘‰๐‘‘| โ‰ฅ 0.1 ๐‘Ž๐‘›๐‘‘ ๐‘‰๐‘  = ๐‘‰๐‘ = 0, transistor forms a weak inversion layer. Transistor in a weak inversion has a constant voltage across the semiconductor channel and the longitudinal electric field across the channel is null. Thus there is no drift current generating inside. Instead the leakage current ๐‘ฐ ๐‘ซ,๐’˜๐’†๐’‚๐’Œ is generated by the diffusion of majority carriers across the channel. We can mathematically model this sub-threshold drain leakage ๐‘ฐ ๐‘ซ,๐’˜๐’†๐’‚๐’Œ with the following factors.
  28. 28. o ๐‘ˆ ๐‘‡ โˆ’ ๐‘‡โ„Ž๐‘’๐‘Ÿ๐‘š๐‘Ž๐‘™ ๐‘‰๐‘œ๐‘™๐‘ก๐‘Ž๐‘”๐‘’ (๐‘ƒ๐‘Ž๐‘Ÿ๐‘Ž๐‘š๐‘’๐‘ก๐‘ค๐‘Ÿ ๐‘™๐‘–๐‘›๐‘’๐‘Ž๐‘Ÿ๐‘™๐‘ฆ ๐‘‘๐‘’๐‘๐‘’๐‘›๐‘‘๐‘’๐‘›๐‘ก ๐‘œ๐‘› ๐‘กโ„Ž๐‘’ ๐‘ฃ๐‘œ๐‘™๐‘ก๐‘Ž๐‘”๐‘’) o ๐‘ฐ ๐ŸŽ โˆ’ ๐ผ๐‘›๐‘–๐‘ก๐‘–๐‘Ž๐‘™ ๐ท๐ถ ๐‘œ๐‘“๐‘“๐‘ ๐‘’๐‘ก ๐‘‘๐‘Ÿ๐‘Ž๐‘–๐‘› ๐‘๐‘ข๐‘Ÿ๐‘Ÿ๐‘’๐‘›๐‘ก It is paramount important to look at the exponential dependency of ๐‘ฐ ๐‘ซ,๐’˜๐’†๐’‚๐’Œ on ๐‘‰๐‘”๐‘  as well as the linear offset based on ๐‘‰๐ท๐‘† ๐‘ฐ ๐‘ซ,๐’˜๐’†๐’‚๐’Œ = ๐‘พ ๐‘ณ ร— ๐‘ฐ ๐ŸŽ ร— ๐’†(๐‘ฝ ๐’ˆ๐’”โˆ’ ๐‘ฝ ๐’•๐’‰)(๐’Ž ๐‘ผ ๐‘ป )โˆ’๐Ÿ ร— (๐Ÿ โˆ’ ๐’†โˆ’๐‘ฝ ๐‘ซ๐‘บ ร—๐’Žร—๐‘ผ ๐‘ป โˆ’๐Ÿ ) 2.2.1.3.1.3 ๐‘ฐ ๐‘ฎ๐‘ฐ๐‘ซ๐‘ณ โˆ’ Gate Induced Drain Leakage Gate-induced drain leakage is generated when a large enough gate to drain ๐‘‰๐‘”๐‘‘voltage is applied to produce a band to band electron tunneling near the interface between the gate oxide and the semiconductor of the drain. 2.2.1.3.2 ๐‘ฐ ๐‘ฎ๐‘จ๐‘ป๐‘ฌ โˆ’ Gate Leakage The gate leakage current is dribbling across the gate to and from the channel, substrate, and diffusion terminal. This current avoid the treatment of the gate of a device as an ideally insulated electrode. This gate leakage is basically composed of two leakage components. o ๐ผ ๐‘‡๐‘ˆ๐‘๐‘๐ธ๐ฟ โˆ’ ๐บ๐‘Ž๐‘ก๐‘’ ๐‘‡๐‘ข๐‘›๐‘›๐‘’๐‘™๐‘–๐‘›๐‘” o ๐ผ ๐ป๐ถ โˆ’ ๐ป๐‘œ๐‘ก ๐ถ๐‘Ž๐‘Ÿ๐‘Ÿ๐‘–๐‘’๐‘Ÿ ๐ผ๐‘›๐‘—๐‘’๐‘๐‘ก๐‘–๐‘œ๐‘› 2.2.1.3.2.1 ๐‘ฐ ๐‘ป๐‘ผ๐‘ต๐‘ต๐‘ฌ๐‘ณ โˆ’ Gate Tunneling This leakage current is generated due to carriers tunneling through the gate of the transistor. There are two major different way of carrier tunneling. ๏‚ท Fowler-Nordheim Tunneling Tunneling into the conduction band of the dielectric. It manifest itself as electron emission caused by the intense high electric field. ๏‚ท Direct Tunneling Tunneling to or from the gate through the forbidden band gap of the dielectric 2.2.1.3.2.2 ๐‘ฐ ๐‘ฏ๐‘ช โˆ’ Hot Carrier Injection
  29. 29. This current is known as Hot Carrier Leakage which is origin whenever a carrier gains enough kinetic energy and overcome the gate potential barrier. This is more often happen to electrons since the voltage barrier and effective mass of an electron is less than the one for holes. 2.3 Technology Scaling and Power Estimation Revolutions 2.3.1 CMOS Device Scaling Although the rate is slowing device scaling is slowing down and deviating from the actual moorโ€™s law, technology sand device scaling is happening so rapidly. Tehgrefore designers have potential of inexpensive doubling the number of transistors every two years, which is possible thank to the miniaturization of devices and the reduction of the cost of computer power due to sales volume. Therfore desifners have to be careful about the stuff because in near future the transistor miniaturization reaches atomic level. Some challenges still remain for the visionars. ๏ƒ˜ Photolithography ๏ƒ˜ Manufacturing Cost ๏ƒ˜ Increased power density ๏ƒ˜ ๐ผ ๐‘œ๐‘› ๐ผ๐ฟ๐‘’๐‘Ž๐‘˜๐‘Ž๐‘”๐‘’โ„ current ratio When we take the leakage current perspective there are positive as well as negative effects. But more oftenly the compact perspective is negative. That means it will increase the power consumption. With this device scaling according to the device physics, some leakage currents such as ๐‘ฐ๐’Š๐’๐’— are to shrink, some leakages such as ๐‘ฐ ๐‘ป๐‘ผ๐‘ต๐‘ต๐‘ฌ๐‘ณ , ๐‘ฐ ๐‘ซ,๐’˜๐’†๐’‚๐’Œ are to be reduced. Device miniaturization and their impact on leakage is very important in the near future. 2.3.1.1 Gate Oxide Thickness (๐‘ป ๐‘ถ๐‘ฟ) Scaling As long as semiconductor device is taking place, the gate oxide thickness and effective channel length is getting reduced. This thicknes is one of the important parameter to FEt and MOS devices where it is directly enguaging with the MOS capacitance and all that. Accordign to some papaers the relationship of the ๐‘ป ๐‘ถ๐‘ฟ scaling can be illustrated as , ๐ฟ ๐‘’๐‘“๐‘“ = 45 ร— ๐‘‡๐‘‚๐‘‹ Where, ๐ฟ ๐‘’๐‘“๐‘“ โˆ’ ๐ธ๐‘“๐‘“๐‘’๐‘๐‘ก๐‘–๐‘ฃ๐‘’ ๐ถโ„Ž๐‘Ž๐‘›๐‘›๐‘’๐‘™ ๐ฟ๐‘’๐‘›๐‘”๐‘กโ„Ž This relationship is usually leds to good ๐‘‰๐‘” โˆ’ ๐ผ ๐‘‘transfer behavior. With this device scaling and other restrictions, Gate Oxide Thickness is limited to some typical level and create a barrier. There are two leakage components which are affected by the ๐‘ป ๐‘ถ๐‘ฟ scaling,
  30. 30. o ๐ผ ๐บ๐ผ๐ท๐ฟ โˆ’ ๐บ๐‘Ž๐‘ก๐‘’ ๐ผ๐‘›๐‘‘๐‘ข๐‘๐‘’๐‘‘ ๐ท๐‘Ÿ๐‘Ž๐‘–๐‘› ๐ฟ๐‘’๐‘Ž๐‘˜๐‘Ž๐‘”๐‘’ o ๐ผ ๐‘‡๐‘ˆ๐‘๐‘๐ธ๐ฟ โˆ’ ๐บ๐‘Ž๐‘ก๐‘’ ๐ท๐‘–๐‘Ÿ๐‘’๐‘๐‘ก ๐‘‡๐‘ข๐‘›๐‘›๐‘’๐‘™๐‘–๐‘›๐‘” ๐ผ ๐บ๐ผ๐ท๐ฟ current is increase, since the voltage required to generate electron tunneling decreases as gate oxide thicknes shrinks. Although the ๐ผ ๐บ๐ผ๐ท๐ฟ could impose a limit on scling the ๐‘ป ๐‘ถ๐‘ฟ , its effect is expected to be less relevant for digital applications such as the voltage reduces below the energy band gap of the silicon. Direct tunneling and hot carrier injection is expected to increase significantly as the thin oxide layer become smaller and smaller. Instead of deployment of Silica as a insulating materials, usage of other dielectric will reduce the gate oxide thickness massively in the near future. 2.3.1.2 Channel Miniaturization As long as technology scalig happens the effective channel is reducing and short-channel effects are expected to worsen when channel length is reduced. DIBL โ€“ Drain Induced Barrier Lowering is one such effect. In DIBL scenario whwre depletion region of the source or drain extends into the channel of a MOSFET device, effectively reducing the channel length. This reduction in depletion layer lowers the potential barrier for electrons, which results in an observable lowering of ๐‘‰๐‘กโ„Ž, and hence in an increase on the ๐‘ฐ ๐‘ซ,๐’˜๐’†๐’‚๐’Œ current. And also channel miniaturization reduces the junction area netween the substrate and the Source or Drain, effectively reducing the ๐‘ฐ๐’Š๐’๐’—. Channel miniaturization is also closely related to scalling of the ๐‘ป ๐‘ถ๐‘ฟ and optimization is paramount importance for the best power performances. 2.3.1.3 Supply Voltage and Threshold Voltage (๐‘ฝ ๐’…๐’… & ๐‘ฝ๐’•๐’‰ ) Scaling ๐‘ฝ ๐’…๐’… and ๐‘ฝ๐’•๐’‰ are two vital transistor characteristics. Design engibeers typically scale the supply voltage ๐‘ฝ ๐’…๐’… to control dynamic power consumption and power density. In the same time reduction of ๐‘ฝ ๐’…๐’… forces a dramatic reduction in Threshold voltage ( ๐‘ฝ๐’•๐’‰) inorder to raise the performance gains. This reduction in ๐‘ฝ๐’•๐’‰ typically causes a relatively large increase in ๐‘ฐ ๐‘ถ๐‘ญ๐‘ญ, while the reduction of ๐‘ฝ ๐’…๐’… reduces the leakage current substantially. 2.3.1.4 Doping Concentration Electric field at a p-n junction strongly depends on the junction doping. As long as technology and device scaling continue, the doping concentration is rising, incrementing the overall ๐‘ฐ๐’Š๐’๐’— and ๐‘ฐ ๐‘ป๐‘ผ๐‘ต๐‘ต๐‘ฌ๐‘ณ. Device engineers smart enough buil a smart doping profiles for the channel and the transistor terminals to maximize active current driving through the channel while minimizing the idle current. 2.3.1.5 Source Drain Punchthrough
  31. 31. Punchthrough happens when the depletion regions from the source and the drain join in the absence of a depletion region induced by gate. Punchthrough happens when voltages between the source and the body are above then nominal range of ๐‘‰๐‘‘๐‘‘, since this is not the common case for digital circuits. Therefore model punchthough is very important to ASIC designers. 2.4 Miscellaneous Sources of Power Consumption Meta stability: output of the flops are becoming undefined Glitches : Unwanted state changes in nets Latch Ups : Creating short circuit paths due to misuse 2.4.1 Metastability Output of the flops are stucked on the undefined states which s caused by the violation of setup time and hold time. ๏‚ท Set Up Time Amount of time that the input signal needs to be stable before clocking the flop ๏‚ท Hold Time Amount of time that input signal wants to be stable after clocking the flop Violating tehse time constraints cause the output od the flop to become unknown and variable for short period of time. After this short time period the output of the flop will fall to 1 to or 0 state regardless of the desired output. 2.4.2 Glitches Glitches are unwanted or undesired changes in signals which are resilient (self correcting). Glitches are caused by delays in lines and propagation delays of cells. Glitches mean more state changes in the signal,thus more switching, internal and leakage power dissipations. Unfortuantely glitches can be propagated through combinational logics cells until being terminated at a flop edge. Therefore whole system will put into error disabled state. Therfore identifying glitch sections and avoid them is paramount important. 2.4.3 LatchUps LatchUps is a short circuit path between supply and the ground. Large amount of power can be dissipated due to latchups.
  32. 32. 2.5 Power Estimation Techniques in Electronic Design Automation (EDA) Electronic Design AutomationEDA Power Estimation Methodologies ๏ต Mostly based on the tech libraries ๏ต Based on two major calculations ๏ต Activity ๏ต The number of toggles per clock cycle on the signal, averaged over many cycles ๏ต Probability n OUT IN ppnn p
  33. 33. ๏ต Percentage of the time that the signal will be high
  34. 34. 3. Power Reduction VLSI power is becoming paramount important to the electronic world because of its emerged significant in both design and consuming stages. With the rapid technology evolution of last couple of decades, increases in transistor density and speeding clock frequency have emerged inherent strength of power architecture of VLSI designs. And also seeking trends of consumers appetites in high performances, portability and embedded applications are clamoring for new power architecture of VLSI designs. We can classified them into three main categories based on that. ๏‚ท Device Engineering ๏‚ท Circuit Engineering ๏‚ท System Engineering Device Engineering This refers to techniques that are implemented on the underlying transistor that form digital circuitry. This is mostly involved with the transistor level components. Circuit Engineering These refer to techniques that are applied to gate/logic level, which are clusters of transistors that perform a small computation like NAND, NOR etc. System Engineering These are referring to techniques that can be applied to macro-blocks that are part of a big data path or micro-chip. 3.1 Dynamic Power Reduction Dynamic power is very important and most contributing power factor for total power consumption. Reduction of dynamic power have already researched and implemented in lots of ways. Thus rather than a radical solutions, improvements and incremental developments are taking place all around the world to mitigate dynamic power. But most of the time dynamic power reduction techniques can be deployed in synchronous electronic circuits which are synchronized with the system clocks. But we know that there is a great potential for them to be used in asynchronous circuit also since switching and internal power dissipations are happening in asynchronous circuits. Dynamic power reduction techniques such as clock gating has emerged its counterpart as advanced dynamic power reduction technique.
  35. 35. 3.1.1 Clock Gating Each time a flop is toggling, immersive switching power and internal power are dissipating out of the semiconductor devices. When we consider millions and billions of such gate level devices large amount of power is wasting due to unwanted usage of toggling and flops. Thus provide a state changes in the intermediate combinational clouds and dissipate more dynamic power. Dynamic power is generated due to unwanted clock toggles, thus charging and discharging internal and load capacitance in internal and switching power dissipation respectively. Changes in the combinational logic is happens only when flop changes its state, thus preventing flop to be toggled , when no state changes are happening is not really effect the switching power of the combinational clouds. Changes of signal state of the combinational clouds are happening only when flop change its state. Local clock gating is creating enable and allow clock to make a change/toggle in flop only where it is necessary. Therefore we stop providing the clock to flop by making an enable. But how we can we really decide enable signal, because itโ€™s not in the real design. Therefore we need to compose a signal with other signals. Having an independent and individual enable is good for simple designs, but highly complex designs compromising with millions of transistors are not sustainable to such implementations. Therefore composing an enable and strengthening them is bit challenge in the high level designs. 3.1.1.1 Local Clock Gating Local clock gating is making enable, and gated it with an actual system clock and, gated clock will be fed as the clock to flop. Thus prevent unwanted toggling of the flop. But local clock gating has no mean unless we have a better algorithm to decide new enable. General implementation of clock gating is as follows.
  36. 36. There are two types of local clock gating. o Latch Based Clock Gating o Latch Free Clock Gating 3.1.1.1.1 Latch Based Clock Gating But this will cause to unwanted glitches in the design. Therefore special clock gating mechanism is used. 3.1.1.1.2 Latch Free Clock Gating AND FLOP D Q Enable Clock Data IN Data Out AND FLOP D Q Enable Clock Data IN Data Out Gated CLK
  37. 37. Dynamic power reduction is very important because, ๏‚ท Clock tree consume more than 50% of dynamic power consumption. o Power consumed by combinational logic whose values are changing on each clock edge o Power consumed by flops o Power consumed by the clock buffer tree ๏‚ท Asynchronous Logic Circuits which is not driven by the global clock, is also changing de to state changes in the flops. For the further explanations, sometimes the flop with a clock gating cell can be compressed as integrated into the flop as follows, AND FLOP D Q Enable Clock CLK Data Out Data D Q FLOP Gated CLK FF D Q Enable Clock Data IN Data OutEN CLK
  38. 38. 3.1.1.2 Advanced Clock Gating Techniques 3.1.1.2.1 Multi-Level Boolean Logic Multi-Level Boolean Minimizations 3.1.1.2.1.1 Satisfiability Donโ€™t Care (SDC) Design spots where certain input/ input combination to a circuit can never occur. There may be possible causes for the SDC conditions. o We can represent node using primary inputs and intermediate variables (๐ต ๐‘›+๐‘š ) o The intermediate values depends on the primary input o Therefore, not all the min-terms of the ๐ต ๐‘›+๐‘š can occur As a example, ๐‘ฆ = ๐‘Ž + ๐‘ , ๐‘กโ„Ž๐‘’๐‘› {๐‘ฆ = 0, ๐‘Ž = 1, ๐‘ = ~}๐‘ค๐‘–๐‘™๐‘™ ๐‘›๐‘’๐‘ฃ๐‘’๐‘Ÿ ๐‘œ๐‘๐‘๐‘ข๐‘Ÿ (๐‘†๐ท๐ถ) SDC is one of widely used techniques in multi level Boolean minimization. 3.1.1.2.1.1 Observability Donโ€™t Care (ODC) Design spots where local changes cannot be observed at the primary outputs. Observability donโ€™t care situations will be created when, ๏‚ท Signals at pre-specified observation points (primary outputs) are outputs from some intermediate gates ๏‚ท Changes of some inputs to the intermediate gates may not change the output ๏‚ท Therefore these changes are not observable As a example, ๐’š = ๐’‚ + ๐’ƒ, ๐’˜๐’‰๐’†๐’ ๐’‚ = ๐Ÿ, ๐’„๐’‰๐’‚๐’๐’ˆ๐’† ๐’๐’ ๐’ƒ ๐’Š๐’” ๐’๐’๐’• ๐’๐’ƒ๐’”๐’†๐’“๐’—๐’‚๐’ƒ๐’๐’† New Trends of Clock Gating with Multi level Boolean logics ๏‚ท Stability Conditions (STC) ๏‚ท Observability Donโ€™t Care (ODC) Conditions Stability Condition has derived through the SDC techniques in Multi level Boolean logics. Amount of power saving depends on the assertion time (duration) of the enable signal. Shorten the enable signal duration, more power saving can be taken place.
  39. 39. 3.1.1.3 Advanced Clock Gating 3.1.1.3.1 Stability Condition (STC) Stability condition is defined with the stability of the input to the flop when upstream flop is stable, no new data or changes come to the downstream flop. If the input to the flop is not changing with the (Stable) for a period of time, there is no use of toggling the flop for state changes. In such situation input to the flop is just remain constant thus output of the flop also stable without changing. Therefore we can shutdown that flop with clock gating. Now instead of pure local clock gating, we have a logic/enable to gate the clock. Then we can reactivate the flop by providing the clock, when a change occurs. ๏ƒผ Advantages: Reduced activity on clock buffer ๏ƒผ Disadvantages: Area and power overhead with the new flop and clock gate Sequential analysis techniques can be applied to apply to find the enable for stability conditions (STC). In this techniques we usually called flop is looking forward, basically upstream with relevant inputs. EN1 Upstream register Downstream register Digital Circuit before STC EN CLK
  40. 40. 3.1.1.3.2 Observability Donโ€™t Care (ODC) There are Situations where the output of the flop is changing or staying constant, but that output is not used in the downstream and read only for a certain time period of time. Then toggling and state changes of the flop for entire time period is not required. Therefore we can shut down that flop for a relevant time period where the output of the flop will not be read and unnecessary. And we can reactivate the flop when someone is actually reading its output. CLK EN Digital Circuit after STC 0 1 Q
  41. 41. Here also we have an area and power overhead and with ODC techniques we can reduce in both clock and data path. 3.1.1.4 Clock Gating Efficiency and Enable Strengthening Enable Strengthening ๏‚ท Most of the devices have explicit or already instantiated clock enables in the digital designs according to records advanced SOC designs such as mobile application units is recommended to have around 90% of clock gating cross designs. ๏‚ท Although the digital designs consist of explicit or instantiated clock enables, all of them are not efficient and provided an efficient clock gating. ๏‚ท Therefore modern approaches are focusing on finding a new enable which strengthen the existing enable. ๏‚ท This process and new enable are often known as Enable Strengthening and the Strengthened Enable respectively. ๏‚ท Basis behind this approach is to strengthen the existing one with new one, if the percentage of power reduction through the new enable surpasses the existing enable. SEL 0 1 Q X CLK SEL 0 1 Q X Digital Circuit After ODC Digital Circuit Before ODC
  42. 42. There are two types of strengthening methodologies based on the logic they are acquired. o Strong STC o Strong ODC Strong STC In a gated flop, if the input is not changing for a period of time and the flop is still clocking or toggling then we can find out a condition for causing input to be stable. We can use this new logic to strengthen the existing enable. Strong ODC In a gated flop, if the output is not read for a period of time but the flop is still clocking, we can find out the conditions for output not t be observed. Then we can enable the existing enable with this new logic. This is known as strong ODC. 3.1.1.5 Memory Power Reduction Most off the digital systems are associated with memory systems. There are different techniques for memory power reduction. o Remove redundant read o Remove redundant write o Memory as steering point for register power reduction o Light sleep power reduction 3.1.1.5.1 Redundant Read Removal Any read access occurring when the memory output is not observable is a redundant read and can be removed based on the ODC technique. And also if the read address is stable then every read after the first one is redundant, if no new address write is taken. This is based on the STC techniques.
  43. 43. 3.1.1.5.2 Redundant Write Removal If the data and write addresses are stable, then ever write access after the first one is redundant and can be removed.
  44. 44. There are techniques which are lightly used in digital gating designs. o Memory as steering point for register power reduction o Light sleep power reduction 3.2 Static Power Reduction In the past few decades dynamic power is the major concern of design engineers due to fastening the system clock and frequency. But prevailing technology revolution with advanced fabrication techniques with technologies such as photolithography, the device or technology scaling is happening with an exponential growth. Thus semiconductor devices scale down and leakages are becoming paramount important for the overall power consumption. Therefore VLSI power architecture predicts that static power (Leakage Power) will become a dominant component of the power architecture and most researches are carrying through to support that concept. Power gating are effectively mitigating leakage losses and becomes a major static power reduction technique. Comparison with Dynamic Power Technology world has predicted that how leakage losses will emerge their counterpart as a major power factor.
  45. 45. Technology scaling is happening rapidly in the recent years; even 10nm level depth has already conquered the semiconductor design world. Bin near future there may be a barrier to grow further because device scaling has already forms restrictions on scaling down from nano scale towards atomic level. We need to have a good understanding about the leakage power leverages to go for new techniques. Emerging Leakage Power in ASIC and SOC (System on Chip) Designs Emerging Leakage Power in Microprocessor Design
  46. 46. Static power reduction techniques are the most influential techniques for the design architecture and VLSI organization. Therefore for analytical purposes we can say that static power gating techniques can be applied in three different levels of the digital designs. Therefore W can classified them into three main categories based on that. ๏‚ท Device Engineering ๏‚ท Circuit Engineering ๏‚ท System Engineering Device Engineering This refers to techniques that are implemented on the underlying transistor that form digital circuitry. This is mostly involved with the transistor level components. Circuit Engineering These refer to techniques that are applied to gate/logic level, which are clusters of transistors that perform a small computation like NAND, NOR etc. System Engineering These are referring to techniques that can be applied to macro-blocks that are part of a big data path or micro-chip. Sources of Static power, basically leakage power has already discussed in the previous chapter. There are two major methods of static power reduction 1. Voltage Islands
  47. 47. 2. Power Gating Although the voltage islands are use for some extent power gating is the most preferred methods for the static power reduction. 3.2.2 Power Gating The basic strategy of power gating is to establish two power modes, ๏‚ท Active Mode ๏‚ท Low Power Mode And switch between these power modes where necessary. Establishment of two power modes is a pragmatic remedy for accurate switch between these modes at the appropriate time and in the appropriate manner to maximize power saving while minimizing the impact on the performance. Therefore switching and controlling process is also complex. ๏‚ท Active Mode ๏‚ท Sleep (Low power mode) ๏‚ท Wake Up Power gating techniques essentially increase the effective resistance of leakage path by adding sleep transistor between logic stack and power supply rails. Transistor stacking techniques are widely concerned in the power gating techniques. Power gating is basically cut of the power where unnecessary. More often these power gating or sleep transistors are shared among the multiple-logic stacks to reduce the number of leakage paths. Sleep transistors are important design of the power gating consideration. Sharing the transistor effectively creates two new power nets, Gated- ๐‘‰๐ท๐ท (๐‘”๐‘ฃ๐‘‘๐‘‘๐‘ฃ) and Gated โ€“Ground (๐‘”๐‘ฃ๐‘ ๐‘ ๐‘ฃ), which replace ๐‘‰๐ท๐ท and GND for power gated logic stack. 3.2.2.1 Sleep Transistors (Switches) 3.2.2.1.1 Switch Types There are two types of sleep transistors. 1. Head Sleep Transistor/Switch 2. Foot Sleep Transistor/Switch
  48. 48. ๐‘”๐‘ฃ๐‘‘๐‘‘๐‘ฃ is connected to ๐‘‰๐ท๐ท using the head sleep transistor. ๐‘”๐‘ฃ๐‘ ๐‘ ๐‘ฃ is connected to GND using a foot sleep transistor. 3.2.2.1.2 Switch Sizing Smaller Switches: Smaller area, large resistance and good leakage reduction Bigger Switches: Larger area, smaller resistance and relatively low leakage reduction 3.2.2.1.3 Switch Placing Architecture Switch in Cell: Switch transistor in each standard cell. Area overhead is a disadvantage and physical design easiness of EDA is an advantage
  49. 49. Grid of Switches: Switches placed in an array across the power gated block. 3 rails routed through the logic block (Power, GND and Virtual). Ring of Switches: Used primarily for legacy design where the physical design of the block may not be disturbed.
  50. 50. 3.2.2.1.1 Signal Isolation Powering Down the region will not result in crowbar current in many inputs of powered up blocks. None of the floating outputs of the power-down block will result in spurious behavior in the power-up blocks. Clams will add some delays to the propagation paths. Basic Power Gating Modes o Fine Grained Power Gating o Coarse Grained Power Gating 3.2.2.2 Power Gating modes 3.2.2.2.1 Fine-Grained Power Gating Process of adding a sleep transistor to every cell is called a fine-grained power gating. This to be turned off imposes a large area penalty. And also individual and independent gating of the power of every cluster of cells form timing issues introduced by inter-cluster voltage variation that are difficult to solve. This technique encapsulates the switching transistor as a part of the standard logic cell. Mostly the gating transistor is designed as the high ๐‘‰๐‘‡. 3.2.2.2.2 Coarse-Grained Power Gating
  51. 51. Implementation of grid style sleep transistor, to stack of logic cell, which drive cell locally through shared virtual power network, is known as coarse grain power gating. There are two ways of doing power gating with coarse-grain approach. ๏‚ท Ring Based o Power gates (Switches) are places around the perimeter of the module that is being switched off as a ring ๏‚ท Column Based o Power gates are inserted within the module with the cells abutted to each other in the form of columns 3.2.2.3 Power Gating Factors ๏‚ท Power gate size ๏‚ท Gate control slew rate ๏‚ท Simultaneous Switching Capacitance ๏‚ท Power gate leakage 3.2.2.4 Synchronous and Asynchronous Power Gating Efficient Power Gating with Asynchronous Digital Design ๏‚ท Clock gating for dynamic power reduction which reduce the power consumption of idle section of synchronous circuits ๏‚ท Asynchronous circuits has a inherent strength of data driven capability and active while performing useful tasks ๏‚ท Asynchronous circuits implement the equivalent of a fine grain power gating network ๏‚ท Power gating can be efficiently implemented in Pipelined flows
  52. 52. 3.2.2.5 Power Gating/Controlling Mechanisms There are two type of Power gating mechanism because of their state preservation mechanism ๏‚ท Non-State Preserving Power Gating ๏‚ท State-Preserving Power Gating o State Retention Registers 3.2.2.5.1 Non State Preserving Power Gating ๏ต Non-State Preserving Power Gating ๏ต Cut-off (CO) ๏ต Multi-Threshold (MTCMOS) ๏ต Boosted-Gate (BGMOS) ๏ต Super Cut-off (SCCMOS) 3.2.2.5.2 State Preserving Power Gating ๏ต State Preserving Power Gating ๏ต Variable Threshold (VTMOS) ๏ต Zigzag Cut Off (ZZCO) ๏ต Zero Delay Ripple Turn On (ZZRTO) ๏ต State Preserving use some retention registers to store states. 3.2.2.5.2.1 State Retention Techniques ๏ƒ˜ Software Based Register Read and Write ๏ƒ˜ Scan Based approach based on using scan chains to store state off chip ๏ƒ˜ Retention Registers Retention Registers When power gating taking place we have to retain some critical register content (FSM State).Saving and restoring state quickly and efficiently is the faster and power efficient method to get the block fully functional after power up. There can be various methods for state retention. ๏ƒ˜ DSP Unit: data flow driven DSP unit can start from reset on new data input. ๏ƒ˜ Cache Processor: This mechanism is good for large residual state retention.
  53. 53. 3.2.2.6 Software/Compiler vs. Hardware Based Power Gating Since power gating is directly impacted on the design organization and architecture there can be two types of power gating. Software/compiler driven power gating and hardware based power gating. Most Researches are ongoing to support those two techniques and most of the time ASIC level/hardware level power gating is emerging its counterpart as best. Basically the basis is on the controlling mechanisms of gating and state preservations. No Power Gating With Gating
  54. 54. Actual Scenario
  55. 55. 3.2.2.7 EDA Power Gating Designp power gating library cells Determine which blocks to power gate Determine state retention mechanism Determine Rush Current Control Scheme Design power gating controller Power gating aware synthesis Determine floor plan power gating aware placement clock tree synthesis Route Verify virtual rail electrical charateristics verify timing
  56. 56. Power Verification Power verification process in the EDA is consisting of the steps of analyzing, monitoring and validating power rules related to EDA tool. Each and every power estimation and reduction rule and algorithms should be test and check against digital cores. It is essential to have verification process in the EDA development cycle e to ensure that the software infrastructure is working properly for electronic prototyping. For the design approach and test case generation we have used HDLs such as Verilog, VHDL etc. Most of the time PERL, C++, and excel is used to analyze the results provided by the EDA flow. As an example Verification of Power Monitors Power monitors where all mathematics is physics come to engineering, and solve a problem of reality. It is Practical problem of a power estimation flow that is solved with power monitors. When we have places where probability of the mathematic is always not true, power monitor comes up with a solution. As a example, ๏‚ท A and B nets have simulation data ๏‚ท C does not have a simulation data According to the mathematical formulas, If A, B are two independent event, A B C AND A B C
  57. 57. ๐‘ƒ( ๐ต ๐ดโ„ ) = ๐‘ƒ(๐ด โˆฉ ๐ต) ๐‘ƒ(๐ด)โ„ If those events are independent ๐‘ƒ( ๐ต ๐ดโ„ ) = ๐‘ƒ(๐ต) = ๐‘ƒ(๐ด โˆฉ ๐ต) ๐‘ƒ(๐ด)โ„ ๐‘ƒ(๐ด โˆฉ ๐ต) = ๐‘ƒ(๐ด) โˆ— ๐‘ƒ(๐ต) In the Above example, ๐‘ƒ(๐ด) = 1 2โ„ ๐‘ƒ(๐ต) = 1 2โ„ โˆด ๐‘ƒ(๐ด โˆฉ ๐ต) = 1 4โ„ Actual Scenario ๐‘ƒ(๐ด โˆฉ ๐ต) = 0 This causes to huge estimated power variations and provide a erroneous values to propagated nets. Power Monitor Solution ๐‘ƒ(๐ด โˆฉ ๐ต) = โˆ‘ ๐‘ƒ(๐ด๐‘– โˆฉ ๐ต๐‘–) ๐‘ 0 Divide the simulation time into the fastest clock slots. And find the probability for each and every portion and integrate them together. A B ๐‘ ๐‘กโ„Ž Slot
  58. 58. Power Fix Power fix is the process of fixing power issues in the design and redesign it using reduced techniques with a background checks. SEC (Sequential equivalence check) is used to navigate and find the effect of associated techniques and new design and compare them against the previous outcomes.

ร—