博文速递：Issues in Physical Design

作者：Feroz Ahmed Choudhary

Issues in Physical Design

Design Parasitics

Parasitic Resistance

·If resistance increases delay also get increases (Delay= R.C)

·As technology shrinks interconnects also shrinks and thus wire resistance will get increase

·To avoid this situation we will increase the height of interconnects

Parasitic Capacitance

·As technology shrinks height of nets getting increase, so sidewall capacitance is increasing

·As technology shrinks the dielectric become thinner, the capacitance will get increases

·To reduce the capacitance, minimize the surface area which can be in common

·So we keep the adjacent metal layers vertical and horizontal in designs

Parasitic Inductance

·Mutual inductance affects: High frequency bus

·Self-inductance affects: Clock nets

·To limit inductance, we provide current return paths for high frequency signals

·Separation and Shielding are the possible remedies

·The rule of thumb has been that when the length of the signal path was long enough to become some percentage of a wavelength that the line itself starts to become a concern for signal integrity

·Prominent above 500MHz & below 130nm for long wire nets & Power/Clock lines

Latch-Up

What is Latch-up?

·Phenomenon occur with CMOS/ BiCMOS circuits

·Generation of a low-impedance path between the VDD supply and the Ground

Reason for Latch-up

·Due to regenerative feedback between the parasitic PNP and the NPN Transistors

Impact in the design

·PN Junctions can produce Parasitic Thyristor

  ·Forms by PNP/ NPN structures

  ·Considerable input current is necessary to activate

·Thyristor formed from parasitic transistors is triggered and generates short-circuit between VDD & GND

·Results in self destruction/ system failure due to the direct connection between VDD & GND

NPN Transistor

·Emitter – drain /source of the N-channel MOSFET

·Base – P Substrate

·Collector – N Well in which the complementary P-channel MOSFET is located

PNP Transistor

·Emitter – drain /source of the P-channel MOSFET

·Base – N Well in which the complementary P-channel MOSFET is located

·Collector – P Substrate

Thyristor/SCR/PNPN diode

·Anode – drain /source of the P- channel MOSFET

·Cathode – drain /source of the N-channel MOSFET

·Gate – P Substrate

Remedies for Latch-up

  Latch-up resistant CMOS process

  ·Reduces the gain of parasitic transistors(use of Si starting material with a thin epitaxial layer on top of a highly doped substrate)

  ·Increase the holding voltage above VDD supply

  ·Increase the dopant concentration of substrate & well (but will lead to higher VT)

  ·Retrograde well structure (Highly doped area at bottom and lightly doped at top)

  Layout techniques

  ·Sufficient space between NMOS & PMOS

  ·This reduces the current gain of the parasitic transistors

  ·limited success because can be increased only to a certain limit

  ·Reduce RS and RW by keeping Substrate & Well contacts as close as possible

  ·Place substrate contacts as close as possible to the source connection of transistors connected to the supply rails (VSS n-devices, VDD p-devices)

  ·This reduces the value of RSUBSTRATE and RWELL

  ·A very conservative rule would place one substrate contact for every supply (VSS or VDD) connection

  ·In Std. Cells based designs a common Well Tap is taking out as per the need

  ·Guard Rings

  ·Gain of transistors is reduced (in analog designs)

Electrostatic Discharge (ESD)

Electrostatic Discharge (ESD)

·When two non-conducting materials rub together, then are separated, opposite electrostatic charges remain on both which attempt to equalize each other

·A transient discharge of static charge that arises from either human handling or a machine contact

Reasons for Electrostatic Discharge

·Thin & vulnerable Gate Oxide of the CMOS makes ESD protection essential for CMOS

·Can be due to inductive or capacitive coupling

·ESD can occur during the removal of extra metal by rubbing in metallization process

·ESD occurs so rapidly that normal GND wires exhibits too much inductance to drain the charge before it can do damage

Impact on the design

·ESD can also burn-out device/ interconnect if thermally initiated

·PMOS is stronger than NMOS in ESD protection, because snap back holding voltage is lower for NMOS

Human Body Model (HBM)

·The actual capacitance of the human body is between 150 pF and 500 pF & the internal resistance of the human body ranges from a few kiloohms to a few hundred

·Peak current ≈ 1.3A, rise time ≈10-30ns

Machine Model (MM)

·MM models the ESD of manufacturing / testing equipment

·Peak current ≈ 3.7A, rise time ≈15-30ns, bandwidth ≈ 12 MHz

·ESD stress caused by charged machines is severe because of zero body resistance

·MM ESD withstand voltage is typically one tenth of HBM

·Most ESD protection circuits can only protect HBM and MM

Charged Device Model (CDM)

·CDM models the ESD of charged integrated circuits

·As more and more circuits and functions getting integrated causes large Die size which provides large body capacitance which in turn stores charges for CDM in the body of IC

·Inductance in the model is mainly due to the inductance of bond wires

·Gate oxide breakdown is the signature failure of CDM stress, in contrast to the thermal failure signature of HBM and MM stress

·CDM stress is the most difficult ESD stress to protect against since fastest transient and has the max. peak current

·Peak current ≈ 10A, rise time ≈1ns

ESD Protection

  The integration of Clamping Diodes

  ·Limits the dangerous voltages and conduct excess currents into regions of the circuit that are safe

  The Protection Diodes

  ·Oriented to be blocking in normal operation

  ·Situated between the connection to the component to be protected and the supply voltage lines safe regions consist primarily of the supply-voltage connections

Electromigration

Electromigration (EM)

·A failure mechanism caused by high energy electrons impacting the atoms in a material and causing them to shift position

·Enhanced and directional mobility of atoms under the influence of an electric field

Reason for Electromigration

·Forms a positive feedback path where EM will cause an atom to move down a wire, slightly narrowing the wire width at that location and increasing the current density

·This increased current density then further increases electromigration, causing more atoms to be displaced Transport of material caused by the gradual movement of ions in a conductor due to the momentum transfer between conducting electrons & diffusing metal atoms

·It is most problematic in areas of high current density

·Significant as size decreases & is most significant for unidirectional (DC) current

Impact in the design

·Excessive EM leads to open (voids) & short circuits (Hillocks) and thus decreases the reliability of the chip

·Approaching life time of device faster

·Increased power consumption

·Higher on-chip temperatures

·High Voltage operation

·High frequency switching

EM Remedies and Precautions

·Wire widening to reduce current density

·Good power management techniques

  ·Bigger Power Grids for power nets (putting power grids on thicker layers)

  ·Wire-widening for signal nets

  ·Better Power Grid planning

  ·Double sizing for Power Greedy nets

·Providing Redundant Vias

·Designing the circuit to run at lower voltage levels

·EM resistance can be increased by alloying with Copper

·Controlling temperature by using a thermal-aware IC design methodology

·DFM techniques that reduce variability

·Besides, need to be aware of “dishing” effect (CMP)

Types of EM checks

Related to Currents

·Average EM checks

·RMS EM checks

·Peak EM checks

Related to Nets

·Signal EM checks

·Power EM checks

·Limits for all these EM checks will be specified in technology file as a function of minimum life of the device, depending on the application

·All the three Current related EM checks need to be satisfied for Signal EM unless otherwise specified

·For Power nets, satisfying Average EM numbers would suffice

EM failure mechanisms

·Timing Failure: Narrowing of the wire will increase wire resistance, which may cause a timing failure if a signal can no longer propagate within the clock period

·Functional Failure: Electromigration will continue until the wire completely breaks, allowing no further current flow and resulting in functional failure

EM Rule Types

·Metal Layer based (This was the only rule used in older technologies)

·Metal length or width dependent EM Rules

·Length and width of upper and bottom Metal and also depends on Via width

·Complex rules with polynomials

Antenna Effect

Antenna Effect

·A phenomenon of charge accumulation in metal segments that are connected to an isolated Gate (Poly) during the metallization process

·This phenomenon occurs during process, so also known Process Antenna Effect (PAE)

·It occurs when conducting net act as antenna, amplifying the charge effect

·The conductive layers are receiving the charge, so termed as Antenna Effect

Reason for Antenna Effect

·Glow discharge during Plasma etching results in electric charging, which when occurred in conductive layer leads to Antenna effect thus termed Plasma- Induced/ Process-Induced damage (PID)

·Charging occurs when conductor layers not covered by a shielding layer of oxide are directly exposed to Plasma

·During process like soldering the chip is protected with some shielding

·But during fabrication there is no such protection & will lead to Antenna effect

·For Aluminium based process PAE is prominent at Etching stage and for Copper based process PAE is prominent at Chemical-Mechanical Polishing (CMP) stage

·If the area of a higher metal layer connected to the Gate through lower metal layer/ layers, then the charge of higher metal layer got added to the lower metal layer which can also cause PAE called Accumulative Antenna Effect

Impact in the design

·If the area of the layer connected directly to the Gate the static charges are discharged through the Gate, the discharge can damage the oxide that insulates the gate and cause the chip to fail

·Fowler-Nordheim (F-N) tunneling current will discharge through the thin oxide and cause damage to it

Remedies for PAE

·Assigning higher metal layers for routing

  ·Higher metal layers will not be connected directly to the Gate Connect various metals through Via connections

·Inserting Jumpers

  ·If PAE is in lower layers then PAE can be reduced by connecting it to higher layers through Jumpers

  ·Jumpers will reduce the peripheral metal length, which is attached to the Gate

·Connecting Antenna diode

  ·If it is in higher layers, Jumper wont be a solution, hence need diodes

  ·As soon as extra charge is induced onto metal/ poly the diode diverts the extra charges to the substrate

  ·But for buffer insertion higher metal layers has to come to lower metal layer (M1 or M2) to connect to pins of buffer and go back and also there may not be enough place for buffer insertion

  ·After routing only we go for antenna check, so Buffer insertion may lead to congestion and DRC violations

Antenna Ratio (AR)

·A design rule to prevent charge accumulation during Metal/ Poly-Si layer etching which limits the area of metal segment connected to the Gate oxide

·Foundries set a maximum allowable AR for the chips they fabricate

·The AR is defined as the ratio of plasma-exposed area As,metal to the gate oxide area Apoly as formulated,

·This rule can be applied to any metal segment connected to the Gate

Antenna Effect possibilities example

·Assume a foundry setting a maximum allowable antenna ratio of 500

·If a net has two input gates that each have an area of 1 square micron, any metal layers that connect to the gates and have an area larger than 1,000 square microns have process antenna violations because they would cause the antenna ratio to be higher than 500

Dominant as technology shrinks

·When oxide thickness reduces

·More metallic structures are added to the chip

Antenna (ANT) Rules

·The Antenna Ratio

·For Aluminium at Etching stage (metal deposition)

  ·The top of the metal is protected by a resist during this step, so the antenna rules for this process should be based on the metal sidewall area

·For Copper at Chemical-Mechanical Polishing (CMP) stage

  ·Charge accumulation occurs during CMP

  ·In this process, the sides of the metal are protected, so the antenna rules need to be based on the metal's top surface area

·Metal used in the process depends on Technology

·From 28nm onwards Aluminium is replacing Copper

PAE as a side effect of the manufacturing process

·Plasma etchers/ ion implanters induce charge into various structures connected to Gate Oxide

·This induced charges destroy the Oxide layer - a permanent damage

·Conductor layer pattern etching processes

·Amount of accumulated charge is proportional to perimeter length

·Ashing processes

  ·Amount of accumulated charge is proportional to area

  ·Ashing processes remove remaining photo resist layers after etching processes of a conductor layer

  ·In the late stage of the processes, the area of a conductor layer pattern is directly exposed to plasma

·Contact etching processes

  ·The amount of accumulated charge is proportional to the total area of the contacts

  ·Contact etching processes dig holes between two conductor layers

  ·In the late stage of the processes, the area of all the contacts on the lower conductor layer pattern is directly exposed to plasma

Crosstalk

What is Crosstalk?

·Refers to a signal affecting another signal being transmitted in vicinity caused by capacitive/ inductive coupling

·Crosstalk is the unwanted coupling of energy between two or more adjacent lines which can change the required signal and is also termed as Xtalk

·Occurs on long adjacent wires

·Can be interpreted as the coupling of energy from 1 line to another via:

  ·Mutual Capacitance, Cm(due to Electric Field)

  ·Mutual Inductance, Lm (due to Magnetic Field)

Impact of Crosstalk in the design

·Functional Failures

  ·Noise induced glitches

  ·If the Glitch duration is that of clock period duration, an extra clock cycle effect

·Timing violations

  ·If aggressor switches in opposite direction to the victim : Setup time Violation

  ·If aggressor switches in same direction to the victim : Hold time Violation

·If the victim line is not terminated at both ends in its characteristic impedance the induced spurious signals can reflect at the ends of the line and travel in the opposite direction down the line

·Thus a reflected near-end crosstalk can end up appearing at the far end and vice versa

Types of Crosstalk

·Energy that is coupled from the actual signal line, the aggressor, onto a quiet passive victim line so that the transferred energy "travels back" to the start of the victim line. This is known as the backward or nearend crosstalk

·Energy that is coupled from the active signal line, the aggressor, onto a quiet passive victim line so that the transferred energy "travels forward" to the end of the victim line. This known as forward or farend crosstalk

Remedies to avoid Xtalk

·Its a 3 dimensional problem, so height, width and length matters

·Noise/Bump violations can be fixed by changing the spacing between critical nets

·Shield the clock nets (critical nets) from other nets by ground lines

·Net Re-ordering

  ·Avoid routing the critical nets parallely for long distances

·Modify the clock net (critical nets) minimum width from normal value to a larger one

  ·This makes the router to skip a grid near clock net to prevent spacing violation

  ·This technique not only reduces crosstalk, but will also have a lower resistance due to larger line width & less side wall capacitance

·Can be fixed either by upsizing (increasing the drive strength) of the victim, or by downsizing (decreasing the drive strength) of the aggressor

Soft Errors

Soft Error (Random Particle Error)

·Soft error is the phenomenon of an erroneous change in the logical value of a transistor, and can be caused by several effects, including fluctuations in signal voltage, noise in the power supply, inductive coupling effects etc., but, majority of soft errors are caused by cosmic particle strike on the chip

·With technology scaling, even low-energy particles can cause Soft Errors

·Soft errors are radiation induced faults which happen due to a particle hit, either by an alpha particle from impurities in packaging material or a neutron from cosmic rays

·When particles strike the silicon substrate they create hole-electron pairs which are then collected by PN-Junctions via drift and diffusion mechanisms

·This collected charge creates a transient current pulse and if it is large enough, it can flip the value stored in the state saving element (bit cell, latch etc.)

·These upsets are called Single Event Upsets (SEU)

Impact in the design

·Soft error can result in incorrect results, segmentation faults, application or system crash, or even the system entering an infinite loop

·When particle strike happens in combinational circuit, the result is a glitch which can then propagate to a latch where it could be clocked in and incorrect data can be latched

Precautions to avoid Soft Errors

·Radiation Hardening: Technique to reduce the Soft Error rate in digital circuits

·Radiation hardening is often accomplished by increasing the size of transistors who share a Drain/ Source region at the node

Self Heating

·If current flows through a wire, then due to the resistance of the wire heat will generate

·Oxide surrounding wires is a thermal insulator, so heat tends to build up in wires

·Hotter wires are more resistive & become slower

·Wire self-heating is only a negligible effect in the supply lines on bulk-CMOS ICs

·Self-heating Design Rule/ Self-heating Limit AC current densities for reliability

  ·Typical limit: JRMS < 1.5 MA/ cm2 (for Aluminum nets)

  ·It limits the unavoidable degradation of Electromigration lifetime due to temperature increase in the current carrying or in any nearby interconnect

本文内容仅代表作者观点，不代表平台观点。

如有任何异议，欢迎联系我们。

往期精彩回顾

2021年的第一场雪！英特尔2020年Q4财报解读

ARM、MIPS、RISC-V三种指令集本质上有何区别？

博文速递：Congestion Analysis

Tcl 与 Design Compiler ：基本的时序路径约束