E3 237 Integrated Circuits for Wireless Communication
Lecture 3: Advanced RF CMOS
Gaurab Banerjee Department of Electrical Communication Engineering, Indian Institute of Science, Bangalore
[email protected]
Outline
•
0.13 um RF CMOS Components
•
An advanced RF-CMOS process
Resistors: Based on Diffusion
N-well resistance (about 400 Ohms/sq)
Diffusion resistance (about 10 Ohms/sq, salicided. 100-200 Ohms/sq, un-salicided) •
N-well and diffusion based resistors are characterized by large parasitics
•
Salicide lowers resistance, good for MOS devices, bad for resistors
•
Salicide blocking is expensive, some processes still do it for specialized poly-resistors.
Resistors: Based on poly-silicon
N+/P+ poly - sheet resistance (about 10 Ohms/sq)
Salicide-blocked poly resistance (about 200 Ohms/sq) •
Low parasitic capacitance, better than diffusion based resistors
•
Some processes customize the poly sheet resistance with salicide blocking – very expensive!
•
Interconnect resistance ( about 100 mOhm/square) and vias can add series resistance
Resistors: Based on MOS Device
•
Device biased in triode –> bias dependent resistor –> nonlinear and PVT variation sensitive
•
Large Parasitic capacitance
•
Large resistance – mega-ohms in tens of square microns
Resistors: Equivalent Circuit Model
•
Series inductor at high frequencies – meander to reduce impact
•
Parasitic capacitance to substrate with dielectric losses
MOS Capacitor
•
Offers the highest capacitance per unit area – 10 fF/ sq. micron
•
Channel length adds series resistance – needs to be minimized for high Q
•
Can be used as a varactor in tuning circuits -> careful, not linear!
•
Requires device to be kept in strong inversion
•
Often used in accumulation mode – n+ diffusion in n well, to avoid keeping the device in strong inversion. The other plate is still the gate.
Metal Capacitors
Lateral Flux Capacitor Metal-Insulator-Metal (MIM) capacitor is like this with special dielectric layer (usually on top of BEOL stack)
Stacked and Inter-digitated Metal-Oxide-Metal (MOM) capacitor is like this with whatever insulator is available in BEOL stack
Aparacio et al, “Capacity limits and Matching Properties of integrated capacitors ”, JSSC, March 2002
•
MOM capacitance per unit area – 1.5-2 fF/ sq. micron
•
High Q and very linear.
•
Symmetric parasitic capacitance – about 10% of total capacitance
•
Accuracy determined by lithography – usually better than 10%
•
Use arrayed unit cells to improve matching.
Capacitors: Equivalent Circuit Model
•
“Lead” inductance and resistance
•
Parasitic capacitance to substrate with dielectric losses – sometimes known as “cap-to-ground” or “bottom plate capacitance”.
Integrated Spiral Inductors
1-turn with underpass
multi-turn, multi-crossover, with center-tap •
Generally uses top metal layer, with or without patterned ground shield
•
Typically considered parameters (L=1-10 nH, Q = 5-10, self resonance frequency = 10-30 GHz )
•
Width, length and area over substrate typically determine losses
•
“Hollow” inductors typically provide better Q for same L – avoid inner turns
•
Several closed form expressions exist –> Start with L = µ0 n2 r, refine with field solvers
•
Usually rounded inductors are slightly more efficient –> does not make a huge difference!
Inductors: Equivalent Circuit Model
•
Series resistance representing metal losses (skin effect) and eddy current losses (substrate)
•
Parasitic capacitance to substrate with dielectric losses
•
Turn to turn and cross-over capacitance – lowers self-resonance frequency
Bond-wire Inductors
•
Generally 1nH/mm (achieves 1-5 nH)
•
Not well controlled, parasitic dominated
•
Q much better than integrated inductors -> around 40
•
Every instantiation requires a bond-pad: difficult for “internal” inductors
Integrated Transformers
A
B
•
Utilizes magnetic coupling between adjacent metal layers.
•
Key Parameters : L, M=Mutual inductance.
•
Typically very lossy, requires tight coupling between primary and secondary.
•
Almost always requires EM field-solver.
•
Studies have shown that “stacked” transformers are more efficient than “planar” transformers.
Outline
•
0.13 um RF CMOS Components
•
An advanced RF-CMOS process
Intel’s 32-nm CMOS for RF
C. H. Jan et al. “RF CMOS Technology Scaling in High-k/Metal Gate Era for RF SoC (System-on-Chip) Applications”, Int. Electron Dev. Meeting, Dec. 2010
•
Performance improvement in modern CMOS processes is not driven by geometrical scaling any more.
•
Advances in materials (strained silicon, high-k gate dielectric, metal gate) are now enabling high fTs, lower noise figures and better figures of merit for RFICs.
fT Scaling Trends to 32-nm
•
Steady improvement in peak fT over the last few generations -> major jump from 90 to 65-nm technology.
•
Strong function of current density -> Burn more current if you need high fT.
Flicker Noise Performance
•
10x reduction in flicker noise from 180 nm to 32 nm technology.
•
Primarily due to increase in gate capacitance, aided by improvements in dielectric materials.
•
Better growth/deposition techniques for advanced processes may have a role to play.
Minimum Noise Figure Performance
•
Steady improvement in NFmin over the last three technology generations.
•
Trend reversal (?) between 45-nm and 32-nm technologies?
•
Generally a tradeoff between improving gm (materials) and parasitics (limitations in fabrication).
Interconnect Scaling
•
Primarily driven by the performance and density requirements of digital processes.
•
Generally, has a negative impact due to interconnect parasitics.
•
In older processes, top metals were deliberately kept thicker for power distribution and RF inductors -> This is getting limited by interconnect scaling.
Thick Metal in Advanced Processes
•
One possible solution -> Add a thick metal layer at the top.
•
Similar to the “redistribution layer” offered by many foundries for RF/Analog.
Key RF Device Characteristics
Next Class: Analog/RF Layout