EE C45 ME C18 Introduction to MEMS Design Fall 008 Prof. Clark T.-C. Nguyen Dept. of Electrical Engineering & Computer Sciences University of California at Berkeley Berkeley, CA 9470 Lecture 7: Noise & Integration ti EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 1
Lecture Outline Reading: Senturia, Chpt. 16 Lecture Topics: Noise MEMS/Transistor Integration Wrap Up Final Exam Next Week EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08
Noise EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 3
Noise Noise: Random fluctuation of a given parameter I(t) In addition, a noise waveform has a zero average value Avg. value (e.g. could be DC current) I D I(t) t We can t handle noise at instantaneous times But we can handle some of the averaged effects of random fluctuations by giving noise a power spectral density representation Thus, represent noise by its mean-square value: Let i ( t ) = I ( t ) I D Then i 1 = ( I I ) = D lim I T T T 0 I D dt EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 4
Noise Spectral Density We can plot the spectral density of this mean-square value: i Δf [units /Hz] One-sided spectral density used in circuits measured by spectrum analyzers Two-sided spectral density (1/ the one-sided) d) Often used in systems courses i = integrated mean-square noise spectral density over all frequencies (area under the curve) EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 5
Inputs Circuit Noise Calculations Deterministic Outputs v i ( jω) ) ( jω) ) H ( jω) S i (ω ) S o o( (ω ) Deterministic: Random: Linear Time-Invariant System Random v π v o (t) ( jω) ω o v o t ω ( jω) H ( jω) v ( jω) o = i v o ω ο S o (t) S ( jω) ) t S o ω ο Mean square spectral density * Random: S ( ω ) = [ H ( j ω ) H ( j ω )] S ( ω ) = H ( j ω ) S ( ω ) o i S ( ω) = H ( jω) S ( ω) How is it we o Root mean square amplitudes i can do this? EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 6 i ω
Handling Noise Deterministically Can do this for noise in a tiny bandwidth (e.g., 1 Hz) S n v n n 1 = S1 ( f Δf ω ο B ( jω) ) ω Can approximate this v = S ( f ) B v n 1 S S o i 1 ( ω ο ω ω o ω [This is actually the principle by which oscillators work oscillators are just noise going through a tiny bandwidth filter] B by a sinusoidal voltage generator (especially for small B, say 1 Hz) v o τ ~ (t) 1 B A cosω t Why? Neither the amplitude nor the phase of a signal can change appreciably within a time period 1/B. EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 7 o t
Systematic Noise Calculation Procedure H ( jω) H5( jω) General Circuit With Wth Several Noise Sources v n i n1 v n3 i n5 i n4 v n6 v on H1( jω) Assume noise sources are uncorrelated 1. For i n1, replace w/ a deterministic source of value in1 n 1 = i Δff (1Hz) EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 8
Systematic Noise Calculation Procedure v 1( ω) i 1( ω) H ( jω). Calculate n (treating it like a deterministic signal) 3. Determine on = v on1 = in1 H ( jω) 4. Repeat for each noise source:,, i n1 1 v n n3. p f,,v n3 5. Add noise power (mean square values) v ontot = v on1 + v on + v on3 + v on4 +L v ontot = v on1 + von + von3 + von4 +L Total rms value EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 9
Minimum Detectable Signal (MDS) Minimum Detectable Signal (MDS): Input signal level when the signal-to-noise ratio (SNR) is equal to unity Sensed Signal Sensor Scale Factor Circuit Gain Output t Sensor Noise Sensor Circuit Output Noise Signal Conditioning Circuit Includes desired output plus noise The sensor scale factor is governed by the sensor type The effect of noise is best determined via analysis of the equivalent circuit for the system EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 10
LF356 Op Amp Data Sheet EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 11
Minimum Detectable Signal (MDS) Minimum Detectable Signal (MDS): Input signal level when the signal-to-noise ratio (SNR) is equal to unity Sensed Signal Sensor Scale Factor Circuit Gain Output t Sensor Noise Sensor Circuit Output Noise Signal Conditioning Circuit Includes desired output plus noise The sensor scale factor is governed by the sensor type The effect of noise is best determined via analysis of the equivalent circuit for the system EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 1
Move Noise Sources to a Common Point Move noise sources so that all sum at the input to the amplifier circuit (i.e., at the output of the sense element) Then, can compare the output of the sensed signal directly to the noise at this node to get the MDS Sensed Signal Sensor Scale Factor Circuit Gain Output Sensor Noise Sensor Circuit Input- Referred Noise Signal Conditioning Circuit Includes desired output plus noise EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 13
r F c Gyro Readout Equivalent Circuit (for a single tine) Noise Sources r r r = mac = m ( x & d Ω ) i f F c l x c x f r x r x η e :1 i o v x i 0 ia C p ia - + R f v Gyro Sense Element Signal Conditioning Circuit Output Circuit (Transresistance Amplifier) Easiest to analyze if all noise sources are summed at a common node EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 14
r F c Gyro Readout Equivalent Circuit (for a single tine) r r r = mac = m ( x & d Ω ) Noise Sources Noiseless l x c x f r r x r x η e :1 i o v F x i 0 c eq C p eq - + R f v Gyro Sense Element Output Circuit Signal Conditioning Circuit (Transresistance Amplifier) v eq i eq Here, and are equivalent input-referred voltage and current noise sources EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 15
r F c = Example: Gyro MDS Calculation (cont) r ma c = l x c x r r m ( x & Ω) f r x r x d η e :1 i o v F x& i v 0 c s eq C p eq - + R f Noiseless Now, find the i eqtot entering the amplifier input: EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 16
Example: Gyro MDS Calculation (cont) EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 17
r F c = Example: Gyro MDS Calculation (cont) r ma c = l x c x r r m ( x & Ω) f r x r x d η e :1 i o v F x& i v 0 c s eq C p eq - + R f Noiseless First, find the rotation to i o transfer function: EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 18
Example: Gyro MDS Calculation (cont) EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 19
Sensing Circuits (cont) EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 0
Problems With Pure-C Position Sensing To sense position (i.e., displacement), use a capacitive load EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 1
F d1 Ele ectrode The Op Amp Integrator Advantage x d 1 d b k e C 1io - m lectrod E The virtual ground provided by the ideal op amp eliminates the parasitic capacitance C p R 1 R >> sc (for biasing) C p 0 + v i C C 1 1 v 1 V P EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08
Integration of MEMS and Transistors EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 3
Merged MEMS/Transistor Technologies (Process Philosophy) MEMS-Last: MEMS-First: Mixed: problem: multiple passivation/protection steps large number of masks required problem: custom process for each product MEMS-first or MEMS-last: adv.: modularity flexibility less development time adv.: low pass./protection complexity fewer masks EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 4
Analog Devices BiMEMS Process Interleaved MEMS and 4 μm BiMOS processes (8 masks) Diffused n+ runners used to interconnect MEMS & CMOS Relatively deep junctions allow for MEMS poly stress anneal Used to manufacture the ADXL-50 accelerometer and Analog Devices family of accelerometers EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 5
Analog Devices BiMEMS Process (cont) Examples: Old New Analog Devices ADXL 78 Analog Devices ADXL-0 Multi-Axis Accelerometer Can you list the advances in the process from old to new? EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 6
50 nm CMOS Cross-Section D D G Sub S nd Level Metal Interconnect (e.g., Cu) G Sub S S G D S G D 1 st Level Metal Interconnect (e.g., Al) LPCVD SiO Polysilicon Gate CVD Tungsten LOCOS Oxidation TiSi Contact Barrier TiN Local P + N P+ N + P N + Interconnect N Well - PMOS Substrate P Well - NMOS Substrate Silicon Substrate P Lightly Doped Drain (LDD) 8 masks and a lot more complicated than MEMS! EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 7
Merged MEMS/Transistor Technologies (Process Philosophy) Mixed: problem: multiple passivation/protection steps large number of masks required problem: custom process for each product MEMS-first or MEMS-last: adv.: modularity flexibility less development time adv.: low pass./protection complexity fewer masks EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 8
MEMS-First Integration Problem: μstructural topography interferes with lithography difficult to apply photoresist for submicron circuits Soln.: build μmechanics in a trench, then planarize before circuit processing [Smith et al, IEDM 95] EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 9
MEMS-First Ex: Sandia s imems Used to demonstrate functional fully integrated oscillators Issues: lithography h and etching may be difficult in trench may limit dimensions (not good for RF MEMS) μmechanical material must stand up to IC temperatures (>1000 o C) problem for some metal materials might be contamination issues for foundry IC s [Smith et al, IEDM 95] EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 30
Bosch/Stanford MEMS-First Process Single-crystal silicon microstructures sealed under epi-poly encapsulation covers Many masking steps needed, but very stable structures Resonator Epi-Poly Seal Epi-Poly Cap Contact Substrate Epi-silicon for CMOS Transistor Circuits Vacuum Chamber [Kim, Kenny Trans 05] μmechanical Device EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 31
Merged MEMS/Transistor Technologies (Process Philosophy) Mixed: problem: multiple passivation/protection steps large number of masks required problem: custom process for each product MEMS-first or MEMS-last: adv.: modularity flexibility less development time adv.: low pass./protection complexity fewer masks EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 3
Berkeley Polysilicon MICS Process Uses surface-micromachinedpolysilicon microstructures with silicon nitride layer between transistors & MEMS Polysilicon dep. T~600 o C; nitride dep. T~835 o C 1100 o C RTA stress anneal for 1 min. metal and junctions must withstand temperatures ~835 o C tungsten metallization used with TiSi contact barriers in situ doped structural polysi; rapid thermal annealing EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 33
Single-Chip Ckt/MEMS Integration Completely monolithic, low phase noise, high-q oscillator (effectively, an integrated crystal oscillator) Oscilloscope Output Waveform [Nguyen, Howe 1993] To allow the use of >600 o C processing temperatures, tungsten (instead of aluminum) is used for metallization EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 34
Usable MEMS-Last Integration Problem: tungsten is not an accepted primary interconnect metal Challenge: retain conventional metallization minimize post-cmos processing temperatures explore alternative structural materials (e.g., plated nickel, SiGe [Franke, Howe et al, MEMS 99]) Limited set of usable structural materials not the best situation, but workable EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 35
UCB Poly-SiGe MICS Process μm standard CMOS process w/ Al metallization P-type poly-si 0.35 Ge 0.65 structural material; poly-ge sacrificial material Process: Passivate CMOS w/ LTO @ 400 o C Open vias to interconnect runners Deposit & pattern ground plane RTA anneal to lower resistivity (550 o C, 30s) Transistor Circuits EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 36
Wrap Up EE C45: Introduction to MEMS Design Lecture 7 C. Nguyen 1/8/08 37