Transcript Modeling and Design for Large Scale Heterogeneous Integration

Modeling and Design for Large Scale
Heterogeneous Integration
Neil Goldsman
Dept. of Electrical and Computer Engineering, UMCP
Collaborators: B. Jacob, A. Akturk, Z. Dilli, T. Chitnis,
M. Khbeis and G. Metze
Modeling and Design for Large Scale
Heterogeneous Integration
Major Goal:
Design, model and fabricate high
performance circuits and systems using
optimized devices, circuits, as well as
3D and wafer-level integration.
Outline
• Conventional Multi-Chip PC Board Integration
 Summary and Limitations
• Solutions
• Research Tasks for Achieving Solutions
 Device and Mixed Mode Modeling




Interconnect Modeling
Circuit Prototyping
Interconnect Prototyping and Characterization
Passive components achievable with 3D fabrication
• Summary for Achieving High Performance 3D Integrated
System
Processor
Integrated Systems
• Systems using a mixed-signal multi-chip PCB architecture:
– Communication circuits
– High performance computers
– Information capture and processing
Input
Output
Limiting Factors in Multi-Chip PCB
Integration
•
•
•
•
Contact Pads
Internal and External Interconnects
Chip Package Pins
Printed Circuit Board Lines and
Connections Between Active Circuits
• Individual Device and Inverter Switching
Speeds and Current Drive
Problems with Conventional Multi-Chip Integration
The parasitics of
the bond pads,
wires and board
buses limit speed,
driving capability
and functionality
Chip-to-chip
PCB
integration
is limiting:
Bond Wire
Bond Pad
Transmission Line
Pins
IC 1
IC 2
Input
Output
Transmission Line
Problems with Conventional Multi-Chip Integration
Increased
demand for
complexity…
Problems with Conventional Multi-Chip Integration
…does NOT
improve
matters
Limitations in Integration
In fact, not only interconnects but each level in the integration
hierarchy imposes its own limitations on the system performance.
•Speed
•Power
•Parasitics
•Speed
•Power
•Noise
•Speed
•Power
•Parasitics (C & L)
•Crosstalk
Search for Solutions to Integration Bottlenecks
•Focus on improved device
design for devices with
higher driving capability on
pads and interconnects
•Focus on improved circuit
design and layout to
minimize the parasitic effect
of pads and interconnects
•3-D integration (3-DI)
•Wafer-scale integration
(WSI)
Wafer Level & 3D Solutions
• To nullify some of the negative effects of chip-tochip interconnects
Wafer-Scale Integration (WSI)
Three-Dimensional Integration
Research Tasks
1.
2.
3.
4.
5.
6.
7.
Model and design devices optimized for speed and drive
Develop mixed signal modeling algorithms to design
fundamental circuit elements for drive and speed.
Develop algorithms and codes for modeling interconnect
parasitics (C and L)
Design and prototype circuits for implementation in 3D and
wafer level integration.
Design and fabricate 3D and wafer level interconnect test
structures and passive circuit elements.
Measure to extract parasitics for characterization of 3D
interconnects.
Combine Tasks 1 through 6 to design and fabricate high
performance 3D circuit.
Task1.
Model and Design devices optimized for
speed and drive
Modeling: Device Dimensions and Doping
Profiles Optimized for Switching & Drive
Asymmetric Doping
1.
2.
3.
4.
5.
6.
Contact Regions: S/D (0.3m wide),
Gate (0.1m wide),
Spacer (0.1m wide)
Gate Oxide (25Å tall, 0.3m wide)
Highly Doped Source/Drain
(0.1m tall, 0.35m wide)
LDD Regions are at the edges
of channel (0.05m tall,
0.05m wide)
Substrate (0.5m tall, 0.9m wide)
Halo (0.01m thick, either around
the Source or Drain)
Doping Profile of a 0.1m Physical Gate Length
Source Side Halo Implanted NMOSFET
DSUB (cm-3)
DHALO (cm-3)
DLDD (cm-3)
DS/D (cm-3)
Halo
1x1016
5x1017
5x1018
1x1020
Conventional
5x1017
-
5x1018
1x1020
Devices
Dopant Concentrations of the Devices
DD Equations
 2  
q
 Si
( p  n  D)
Modeled DC Current Densities for
Different NMOS Structures
Asymmetric Best
n 1
 .J n  Rn  Gn
t
q
p
1
  .J p  R p  G p
t
q
Supplementary DD Equations
J n  q n n  q nVT n
J p  q p p  q pVT p
Coupled Discretized DD Equations are
solved at each mesh point
DC Current Densities of NMOS for
|VGS|=(0.4,0.7,1.0)V and VSB=0V
Electric Fields in Channel Explain
Improvement of Asymmetrical Device
(a) Lateral and (b) Vertical Surface Electric
Fields of NMOS at VGS= VDS =1V
Modeled NMOS Turn-on Characteristics
(a) Turn-on and (b) Turn-off Characteristics of NMOS
Task2.
Develop mixed signal modeling
algorithms to design fundamental circuit
elements for drive and speed.
Mixed-Mode Simulator
DD Equations
 2  
q
 Si
( p  n  D)
n 1
 .J n  Rn  Gn
t
q
p
1
  .J p  R p  G p
t
q
Supplementary DD Equations
J n  q n n  q nVT n
J p  q p p  q pVT p
Coupled Discretized DD Equations are
solved at each mesh point
CMOS Inverter (CMI)
Lumped KCL equation check at the output node
and using the KCL equation, the output guess is
updated for the next iteration, VOi+1:
I DN  I DP  I RL  I CL  0
Voi 1  VSS
Voi 1  Voi
AN  B V  AP  B V 
 CL
0
RL
t
i 1
N o
i 1
P o
RLCL
 ( AN  AP ) RL
t
Voi 1 
RC
1  L L  ( BN  BP ) RL
t
VSS  Voi
Modeled Transfer
Characteristics of CMOS
Inverters
Transfer Characteristics of CMOS Inverters
with(out) a Resistive Load.
Output Current Densities of CMOS Inverters
Switching Characteristics of CMOS
Inverters
Switching Characteristics of CMOS Inverters
without a Capacitive Load.
Switching Characteristics of CMOS Inverters
with Capacitive Loads
Switching Characteristics of CMOS Inverters
with CL=10fF/20m.
Switching Characteristics of CMOS Inverters
with CL=0.3pF/20m.
Net Charge Density During
Low-to-High Input Transition
The variation of the net charge density at the conventional MOSFET during low-to-high transition.
NMOS and PMOS channels are around 0-0.1 and 1.0-1.1 μms along the lateral x, respectively.
Task 3:
Calculation of the Induced Voltage on
Floating Metals via Capacitive Effects
A Typical Bus allows one Write and certain
number of Reads through a single line while
the remaining connections are kept in high
Impedance or floating states.
Voltages induced on the floating pins due to the capacitive
network
Simulation
We developed a simulation tool to find the induced voltages
on the floating metals due to the pins tied up to a fixed
voltage source in an arbitrary rectangular geometry.
 2  0
2
 2 x   y  2 z


0
2
2
2
x
y
z

V
V
dV    S dS  0
S
 S   Ins EIns   Met EMet   ins E Ins
 M  VApplied
Net Charge is zero
inside the metals and the
insulator that is filling the
region between the metals
Total charge concentration on the
floating metals is zero
The potential is constant
for the fixed metals
2D Simulation Result
Here, there are 36 metals uniformly distributed on our mesh grid.
The voltages on the two metals at the opposite corners are set to
0 and 10V, and the rest is left to float.
Potential Distribution
3D Simulation Result
The previous distribution is kept for 3 layers
in the z direction (3rd-5th), and no other
metals are put outside this z range
Potential Distribution at z=4
Isopotential Surfaces
Capacitances In Multiconductor Systems
Q1  C11V1  C12V2  ...  C1NVN
Q2  C21V1  C22V2  ...  C2 NVN
:
The relation between
potential and charge
is linear
QN  C N 1V1  C N 2V2  ...  C NNVN
C ji 
Qi
Vi
V j 0, j i
 C11
C
 C   21
 :

C N 1
C12
C22
CN 2
... C1N 
... C1N 
: 

... C NN 
By LU decomposition, we find the voltages on the floating metals, (VF1-VFL), using the
C matrix and the voltages of the fixed metals, (VM1-VMM).
CF1F1VF1  CF1F2VF2  ...  CF1FLVFL  CM1M1VM1  CM1M 2VM 2  ...  CM1VM M
CF2 F1VF2  CF2 F2VF2  ...  CF2 FL VFL  CM 2 M1VM1  CM 2 M 2VM 2  ...  CM 2VM M
:
CFL F1VF1  CFL F2VF2  ...  CFL FLVFL  CM M M1VM1  CM M M 2VM 2  ...  CM M VM M
Then, we find the potential distribution by the utilization of the Conjugate Gradient Method.
HL
Typical Inductance of Internal Interconnect
t
Self Inductance
w=3um
nH mm
0.35
0.3
0.25
0.2
0.15
0.6
0.8
HL
1
1.2
1.4
Thickness
um
1.6
Proposed Model:
Coupling Maxwell Eqns. to Semiconductor Transport Eqns.:
B
 E  
t
n
   (nV )  0
t

 B 
r E
c t
2
 J

dnme V
nme V
 qn E  (V  B )  nkT 
dt
m
J  neV
Assuming isothermal condition: T= constant.
Collisional time  m 
me
u n , un: electron mobility
e
Nonlinearity: collisional term (mobility) is a nonlinear function.
Keeping inertial term in the momentum equation to account for the
electron finite response time.
Task 4:
Design and prototype circuits for
implementation in 3D and wafer level
integration.
Prototype Processor
PC Update
Program Counter
ADDR
INSTRUCTION
MEMORY
MUX
INSTRUCTION
TGT
RDATA
REGISTER
FILE
DATA
MEMORY
TGT
SRC1
SRC2
OPCODE
SRC1
SRC2
WDATA ADDR
CONTROL
ALU
ALU Result Bus
• Verilog HDL
• Teaching ISA
• TI ‘C6000 DSP ISA
in progress
• Fully pipelined
• Handles interrupts
precisely
• Can boot an RTOS (or
OS …)
Phase-Locked Loops
• Can be designed and operated both for analog and digital signals
FM Transceiver
The quintessential analog communication system representative
Transmitter
Input
Signal
Voltage
Adder

DC
Bias
Receiver
Antenna
Matching
Network
PA
VCO
RC
Network
FM Modulator
Antenna
Mixer
Matching
Network
LNA
IFA
Crystal
Filter
Local
Oscillator
PLL
FM
Demodulator
AA
Output
Signal
Task 5:
Design and fabricate 3D and wafer level
interconnect test structures and passive
circuit elements.
On-Chip Inductors and Transformers:
Surface versus 3D
Current Sheet Approach
•
•
Valid for geometries with small spacing between between the metal lines.
Approximates a square Spiral with four trapezoids forming a square.
•
Adjacent trapezoids don’t have mutual inductance since current flow in them
is perpendicular.
Different Inductor Geometries
Hexagonal Inductor
Square Inductor
Octagonal Inductor
Square Transformer – One Inductor Outside Another
Advantages :
-High Self Inductance
-Low terminal to substrate
capacitance
-Low terminal to terminal
capacitance
Disadvantages :
-Low Mutual coupling
(0.3 – 0.5)
Parameters :
Li = 2.3 nH ( Inner L)
Lo = 3.56 nH ( Outer L)
LT = 9.87 nH ( Total L)
M = 2.00 nH ( Mutual L)
K = 0.35 ( Coupling )
Area = 350 um2
Reducing Parasitic Capacitance of Surface
Inductor
SiO2 supports
Layout
Air Gap formed
by etching
3-D Picture (not to scale)
(Connection to pads not shown)
Removing the Silicon Dioxide reduces the capacitance to substrate and is
expected to improve Inductor Quality Factor.
3D Allows Much Larger L
Vertical Vias
Wafer 2
Wafer 1
Layout
Area of Via = 2 m x 2 m
Dimension of Layout : W = 250 m, L = 450 m
Number of turns = 100 ; Inductance = 115 nH
Air Gap(formed by etched
out Silicon Dioxide)
3-D Picture ( not to scale)
(connection to Pads not shown)
Comparison of Areas and Inductances
Area
32400 m2
48000 m2
Inductance
2.5 nH
8.1 nH
Hexagonal Spirals
32500 m2
48000 m2
2.3 nH
7.6 nH
Octagonal Spirals
35000 m2
55000 m2
3.5 nH
11.7 nH
3-D Inductors
(Expected Inductance)
28800 m2
57600 m2
115200 m2
28.8 nH
57.74 nH
115 nH
Square Spirals
Task 6:
Measure to extract parasitics for
characterization of 3D interconnects.
Preliminary analysis indicates pad, pin
parasitics are significantly detrimental.
Test structure layout complete; submission to
Northrop Grumman immenant. (M. Khbeis,
G. Metze)
Conclusion
• Conventional Multi-Chip PC Board Integration
 Summary and Limitations
• Solutions
• Research Tasks for Achieving Solutions
 Device and Mixed Mode Modeling




Interconnect Modeling
Circuit Prototyping
Interconnect Prototyping and Characterization
Passive components achievable with 3D fabrication
An IC with Bonding Pads
Bond Pad
Die
(Integrated Circuits)
Electro-Static Discharge
(ESD)
Standard Planar Technology Implementation:
IC Chips on PCBs
Chip-to-Chip Connection on PC Board:
PC Board
Bond Pad
Die (Integrated Circuits)
Input
Output
Pins
Bond Wire
Transmission Line
Signal Path in Planar Chip-to-chip Connection
Bond Wire
Bond Pad
Pins
Input
Output
ICs
ICs
Transmission Line
Pad Capacitance
Metal
Insulator
Oxide
Pad Parasitic Capacitance vs. Process
(Best Case assuming top metal, no ESD)
Si Substrate
Plate Capacitor
Metal Layer
Substrate
C pad
Pad Capacitance (fF)
160
140
120
100
80
60
40
20
0
1.5u
0.5u
0.35u
Process
0.25u
Conventional Bonding: Limitations on System
Performance
Digital Circuit Speed
Circuit Driving Capacity
Input Capacitance of A CMOS Inverter:
Example Circuit: Ring Oscillator
NMOS: L W  0.24m  0.48m
Bond
Pad
Cg ,nmos  0.7 fF
Faster!
PMOS: L W  0.24m 1.92m
C pad  40 fF
Slower!
Cg , pmos  2.8 fF
 C g ,total  4 fF
Oscillating Frequency (GHz)
Bond
Pad
2
C pad  10C g ,total
1.5
1
Driving a bond pad is equal to driving TEN inverters!
0.5
No Pads
1 Pad
2 Pads
Number of Pads in Loop
3 Pads
IC Package Parasitics
Bond
Pad
Bond
Wire
Requires impedance
matching in design to:
Pin
• Maximize Power Transformation
• Eliminate Signal Reflection
• Minimize Clock Skew
Bond Wire & Pin
(Several nanoHenry)
Bond Pad
(Several Hundred
femto-Farad)
IC 1
Input
IC 2
Output
Transmission Line
Tune to Z0
Z0
Tune to Z0
Tune to Z0
Effect of Pad and Package Parasitics on
Circuit Operation Speed
Test circuit: Binary-to-Gray-to-Binary converter
Objective: To find out max. clock frequency for which Input = Output
Effect of Pad and Package Parasitics on
Circuit Operation Speed: Simulation Results
Device
Size
( L = 0.25u)
W = 0.5u
W = 1.0u
W = 1.5u
W = 2.0u
Separate
Chips
As a stack
Without pads and pins
Speed
80 MHz
100 MHz
2.5316 GHz
Power
0.98 mW
0.435 mW
2.31 mW
Speed
133 MHz
166 MHz
2.5974 GHz
Power
1.825 mW
1.06 mW
4.5 mW
Speed
200 MHz
250 MHz
2.6316 GHz
Power
2.825 mW
1.757 mW
6.575 mW
Speed
222 MHz
333 MHz
2.6666 GHz
Power
3.425 mW
2.675 mW
8.5 mW
Research into Mixed-Signal
Architectures
• Circuit-level research aims to identify
bottlenecks and limitations more precisely
• Design, simulate and fabricate mixed-signal
circuits for the purpose
–
–
–
–
FM transceiver
PLL
Microprocessors
Image sensor systems
Processor? Insert material from Dr. Jacob
here?
Intra-chip interconnect problems
• Signal networks within the ICs themselves also require
investigation
• Example: Clock and power networks in CPUs
Clock networks: H-networks.
Power networks: Grids.
Reminiscent of some antenna configurations:
EM-interference problems?
• Fan-out and bus-driving problems: Tie-in to new device
and circuit architectures
Current Sheet Approach (contd..)
• We require expressions for Self Inductance of a Trapezoidal sheet and
mutual inductance between two parallel trapezoidal sheets.
• Self Inductance of a Trapezoidal sheet and mutual inductance between
two parallel trapezoidal sheets is calculated as an extension of the
mutual inductance between two unequal sized parallel lines.
• A further extension of this yields expressions for the inductance of a
Square Spiral.
• This can be extended to Hexagons and Octagons.
• For example in a Hexagon the adjacent trapezoids have a component
of mutual inductance.
Current Sheet Approach (contd..)
A general expression for the inductance is
L
Where

n
davg
c

w
s
l
n 2 d avg c1
2
c
[ln( 2 )  c3   c4  2 ]

= Magnetic Permeability
= No. of turns
= Average length of element
= Constants depending on the shape of the inductor
= Fill Factor = (nw + (n-1)s)/l
= Width of conductor
= Spacing between conductors
= Average length of element ( Different from davg for shapes
other than square)