The Race to Replace Quartz

Transcription

1 The Race to Replace Quartz Michael S. McCorquodale, Ph.D. Founder and Chief Technical Officer, Mobius Microsystems, Inc. Berkeley Wireless Research Center, Berkeley, CA 12:30PM February 2, 2007

2 Overview Quartz applications and specifications Why replace quartz? Why not? Emerging technologies Si MEMS FBAR High-Accuracy Ceramic RF-TCHO : Mobius Microsystems Measured performance data RF-TCHO technology Motivation Architecture Conclusions 2of 73

3 Quartz Applications and Specifications 3of 73

4 Quartz Applications and Specifications Timing Every synchronous semiconductor component requires a clock to operate Carrier synthesis RF systems require precision frequency references for carrier frequency synthesis Belkin Bluetooth/LAN USB Print Server USB XTAL clock reference Ethernet XTAL clock reference Processor XTAL clock reference Bluetooth radio XTAL reference (on flip side) 4of 73

5 Most Relevant Metrics Frequency and time domain performance Short-term frequency stability: Jitter and phase noise Total frequency accuracy: Accuracy and precision over Manufacturing process (P) Drift over voltage (V) Drift over temperature (T) Long-term stability or aging (A) Start-up latency, rise/fall time, etc. Environmental performance Sensitivity to microphonics Storage lifetime/degradation Cost Fabrication process technology Production trimming requirements Packaging requirements 5of 73

6 The Show-Stopper Metrics for New non-quartz Technologies Accuracy and stability are most significant Nearly all timing and frequency generation standards have accuracy and stability requirements Other deficiencies may be addressable once sufficient accuracy and stability are demonstrated Other benefits may encourage adoption Reliability Form factor Cost 6of 73

7 Common Interface Protocol Reference Rate and Accuracy Requirements Protocol (Application): Rate ± Required Accuracy CAN/LinBus (Auto): ~khz ±1500ppm ±15kppm USB 2.0 (PC and CE): 48MHz ±500ppm SATA Gen. 1 Gen. 3 (HDD): 25MHz ±350ppm PCI/PCIe (PC): 33/66MHz ±300ppm Embedded µp (PC): ~100MHz ±100 ±300ppm Firewire/IEEE1394 (PC and CE): MHz ±100ppm Ethernet (Data comm.): 50MHz ±25ppm Observations Most rates < 100MHz Due to power on PCB Fundamental physical limit to XTAL frequency based on geometry Most accuracy requirements > ±100ppm 7of 73

8 Observations What accuracy does quartz really provide? ~±50ppm initial error ~±15ppm insertion error ~±15ppm TC ~±10ppm synthesis error ~±10ppm aging for 5 yrs. Total ~±100ppm Higher accuracy requires expensive TCXO Not a surprise that most interface protocols for CE are less accurate than ±100ppm 8of 73

9 Common Radio Reference Frequencies and Accuracy Requirements Protocol (Application): Freq. ± Required Accuracy Bluetooth, Zigbee (Network radios): 20MHz ±25ppm GSM, etc. (Cellular radios): 13MHz ±5ppm ASK TPMS (Auto): 9.838MHz ±238ppm Observations Most frequencies <20MHz Most accuracy requirements <±25ppm Carrier spacing w/ adjacent channels: accuracy must be high Carrier spacing w/o adjacent channels: accuracy relaxed General Observations Clock/timing generation: >±100ppm at <100MHz Carrier synthesis: <±25ppm at <50MHz 9of 73

10 Why Replace Quartz? Why not? 10 of 73

11 Why replace quartz? Benefits to eliminating XTALs/XOs in systems Reduced cost Reduced form factor and PCB footprint Reduced time to market Reduced start-up latency (possibly) Reduced EMI (possibly) Increased reliability Increased integration (opportunity for multiple instances) Quartz is one of the last great hold-outs for microelectronic integration 11 of 73

12 Why not? Quartz is robust and relatively cheap Simple and proven no brainer technology Historical traction Economies of scale with handsets continues to drive cost down to <$0.15/unit Supply chain reliability Volume manufacturability The winner of the race will need to contend with these formidable barriers to entry 12 of 73

13 Emerging Technologies 13 of 73

14 Capacitively-coupled µresonators Surface micromachined poly-si structures with capacitive actuation Benefits Very high-q (>10,000) demonstrated Likely low-power due to high-q Challenges High motional impedance (>kω) Nonlinear transduction causes flicker noise upconversion in oscillator circuits Power handling limits Specialized packaging required Process difficult to integrate with CMOS Frequency trimming required Moderate temperature coefficient Aging (material fatigue) Microphonic sensitivity may be high Status Samples available from Discera Surface Micromachined Si MEMS Clamped-clamped beam poly-si microresonator [Nguyen, McCorquodale, et al.] Disk poly-si microresonator [Nguyen, et al.] 14 of 73

15 Surface Micromachined Si MEMS Piezoelectrically-coupled µresonators ZnO film couples actuation to surface micromachined poly-si beam Remainder of device identical to previous microresonator Benefits Much lower motional resistance than previous µresonator (~100Ω) Same as remaining benefits for previous µesonators Challenges Same as remaining challenges for previous µresonators Status Research area No commercialization effort yet Sense Electrode ZnO Film Drive Electrode Tuning Capacitor Device Layer Oxide Handle Layer Piezoelectric microresonator [Ayazi, et al.] 15 of 73

16 Bulk Micromachined Si MEMS Capacitively-coupled microresonators Bulk micromachined Si structures with capacitive actuation Benefits Bulk technology enables hermetic packaging under CMOS Same as remaining benefits for previous µresonators Challenges No CMOS over MEMS area (cost) Same as remaining challenges for previous µresonators Status Samples available from SiTime Limited volume production CMOS over MEMS still not in production Bulk microresonator [SiTime] Stacked die assembly [SiTime] 16 of 73

17 Film Bulk Acoustic Wave Resonators (FBAR) Piezoelectric FBAR Similar to a quartz XTAL, but a film of piezoelectric material over a Si substrate Benefits High-Q Very low motional impedance No specialized packaging required Challenges Some challenges to integrate with CMOS Low accuracy because film thickness sets frequency Status Now in high-volume production at Agilent for filter products Some research oscillator work with VCOs (Berkeley), but not reference oscillators Drive Electrode Thin Piezoelectric Film Sense Electrode FBAR [Ruby, et al.] 17 of 73

18 High-Accuracy Ceramic Resonators High-Accuracy Ceramic Resonator Ceramic, as opposed to quartz, resonators Benefits Cheaper than Quartz More reliable than quartz, particularly at high-t Very common in automotive for CAN/LinBus Challenges Initial accuracy and aging compromise total frequency accuracy TC also compromises total frequency accuracy Only cost benefit over quartz still a macroscopic device Status ±500ppm samples available from Murata though aging likely puts part out of spec. typical closer to ±3kppm Target application is HS-USB making ceramic a true quartz replacement tech. if ±500ppm can be achieved 18 of 73

19 Radio Frequency Temperature-Compensated Harmonic Oscillator (RF-TCHO ) RF-TCHO All-CMOS temperature compensated harmonic (LC) reference oscillator Benefits All CMOS (lowest cost, size, etc.) Trivial to integrate with host Suitable for harsh environments Already achieves sufficient accuracy Challenges Production frequency trimming Achieving <100ppm accuracy Power dissipation higher in some apps. Status In volume production as IP for USB from Mobius Microsystems Component samples available in Q1 07 from Mobius Microsystems 12MHz USB Macro [McCorquodale, et al.] 19 of 73

20 Measured Performance Data 20 of 73

21 Measured performance for Performance Data CTS 24MHz quartz can oscillator Ecliptek (SiTime) 25MHz Si MEMS oscillator Abracon 12MHz ceramic oscillator Mobius 12MHz RF-TCHO Measured parameters Total frequency accuracy RMS period jitter Phase noise at 10k/100k/1MHz offset from carrier Power dissipation 21 of 73

22 CTS 24MHz Quartz Resonator Technology and Architecture CTS AT-cut quartz crystal Mated to CMOS reference oscillator in can Performance Measured accuracy: ~±10ppm Measured RMS period jitter: 8.19ps Measured phase noise -102/-124/-140dBc/Hz 22 of 73

23 CTS 24MHz Quartz Resonator: Accuracy Frequency Error of AT-Cut XO Frequency Error (ppm) Temperature ( C) 23 of 73

24 CTS 24MHz Quartz Resonator: Period Jitter 24 of 73

25 CTS 24MHz Quartz Resonator: Phase Noise 25 of 73

26 Ecliptek 25MHz Si MEMS Oscillator Technology and Architecture Bulk micromachined Si MEMS resonator stacked and bonded on CMOS (SiTime) Low frequency µresonator + Σ -Ring-PLL Performance Measured accuracy: ±25ppm Measured RMS period jitter: 17.69ps Measured phase noise (@10k/100k/1M): -75/-85/-117dBc/Hz 26 of 73

27 Ecliptek 25MHz Si MEMS Oscillator: Accuracy Frequency Error of Si MEMS Oscillator Frequency Error (ppm) Temperature ( C) 27 of 73

28 Ecliptek 25MHz Si MEMS Oscillator: Period Jitter 28 of 73

29 Ecliptek 25MHz Si MEMS Oscillator: Phase Noise 29 of 73

30 Abracon 12MHz Ceramic Resonator Technology and Architecture Ceramic resonator Mated with Cypress CMOS reference oscillator Performance Measured accuracy: ~±3200ppm Measured RMS period jitter: 8.96ps Measured phase noise -110/-129/ of 73

31 Abracon 12MHz Ceramic Resonator: Accuracy Frequency Error of Ceramic Resonator Frequency Error (ppm) Temperature ( C) 31 of 73

32 Abracon 12MHz Ceramic Resonator: Period Jitter 32 of 73

33 Abracon 12MHz Ceramic Resonator: Phase Noise 33 of 73

34 Mobius 12MHz RF-TCHO Technology and Architecture All-CMOS RF temperature compensated harmonic (LC) oscillator Performance Measured accuracy: ±225ppm Measured RMS period jitter: 7.98ps Measured phase noise -96/-124/-141dBc/Hz 34 of 73

35 Frequency Error of RF-TCHO Mobius 12MHz RF-TCHO : Accuracy Frequency Error (ppm) Temperature ( C) Nominal VDD VDD-10% VDD+10% 35 of 73

36 Mobius 12MHz RF-TCHO : Period Jitter 36 of 73

37 Mobius 12MHz RF-TCHO : Phase Noise 37 of 73

38 Performance Comparison Variable/Metric 24MHz XO 25MHz Si MEMS 24MHz Ceramic 12MHz RF-TCHO Total accuracy (ppm) ~±10 ~±25 ~±3200 ~±225 SSB phase noise (dbc/hz) -102/-124/ /-85/ /-129/ /-124/-141 RMS period jitter (ps) Power 38 of 73

39 Performance Comparison Variable/Metric 24MHz XO 25MHz Si MEMS 24MHz Ceramic 12MHz RF-TCHO Total accuracy (ppm) ~±10 ~±25 ~±3200 ~±225 SSB phase noise (dbc/hz) -102/-124/ /-85/ /-129/ /-124/-141 RMS period jitter (ps) Power 39 of 73

40 Thoughts on Measured Data Quartz is lower power because signal is directly synthesized (no PLL) Si MEMS Phase noise is high due to high loop multiplication factor (low frequency µresonator) and Ring-PLL Power is higher due to PLL and 2 µresonator TC architecture RF-TCHO Phase noise and jitter are competitive with quartz/ceramic how does it work? Power is higher, though competitive why? Accuracy is sufficient for most clock applications 40 of 73

41 RF-TCHO Technology 41 of 73

42 System Observations Current XTAL-replacement work focuses too heavily on component-q Component-Q is compromised by frequency multiplication Component-Q only affects reference oscillator performance Component-Q is only loosely related to jitter High component-q increases start-up latency However, high component-q may imply lower power, though that lower power may be lost in PLL Should consider metrics relevant to the output signal, not the reference signal or reference device Jitter (period, cycle-to-cycle, long-term) Phase noise Frequency accuracy/precision Start-up latency Reliability 42 of 73

43 System Observations Phase and frequency are related by a linear operator dφ ω = dt Frequency mult./div. results in phase noise mult./div.: v n ( t) = V cos( ω t + φ ( t)) o o n v n, mult ( t) = Vo cos( Nωot + Nφn( t)) Using narrowband FM approximation: N P, mult./ div. N = P o o ± o f m f m o log( N Linear freq. trans. results in quadratic change in noise power 2 ) 43 of 73

44 System Observations σ RMS = 8 S ( f ) sin 2 π φ ω 2 m 0 o ft o df m The relationship between phase noise and period jitter (σ RMS ) ω o = fundamental radian frequency T o = fundamental period f m = offset frequency from fundamental S φ (f m )= phase noise at offset f m from fundamental Key observations Phase noise is masked by a trigonometric function with period T o /2 Far-from-carrier phase noise contributes significantly to σ RMS 44 of 73

45 System Observations Phase noise PSD (dbc/hz) N Reference N Decrease noise with freq. division +20log 10 (N) -20log 10 (N) f m (Hz) Component-Q of the reference is degraded by frequency multiplication Frequency division can enhance a low component-q reference Can introduce the concept of an effective Q or an output Q which accounts for frequency translation 45 of 73

46 System Observations Phase noise PSD (dbc/hz) +20log 10 (N) XO reference PLL ring VCO (unlocked) PLL loop BW PLL output path Period jitter integration mask f m (Hz) Reference signal component-q matters only within the PLL loop BW The ring VCO has high far-from-carrier phase noise so jitter is high Remember: σ RMS = 8 2 Sφ( f ) sin π 2 m ftodf ω 0 o m 46 of 73

47 System Observations Summary Frequency div./mult. can improve/degrade signal-q of output signal PLLs with high loop multiplication factors have severely degraded jitter, despite the high component-q reference LCOs have low component-q but division can improve signal-q Far-from-carrier phase noise is a significant contributor to jitter Far-from-carrier phase noise in ring PLLs is very high LC-VCOs have low far-from-carrier phase noise In an LC-PLL, low jitter performance originates from the LC-VCO, not the high component-q reference Component-Q of the reference is marginally important to relevant metrics Effects above dominate signal integrity These effects can be exploited to introduce RF-TCHO Still implies low power, though must add power of PLL 47 of 73

48 RF-TCHO Architecture Architectural concept Free-run an LCO at RF and compensate for temperature, bias, etc. Frequency-divide by a large ratio Architecture ensures low jitter, low phase noise Architecture enables low start-up latency Challenges Initial frequency accuracy Maintaining frequency accuracy via compensation for bias and temperature variation as well as aging Maintaining low power 48 of 73

49 RF-TCHO Architecture Production Trimming Logic Digital Control Process Comp. Bias Stability Sustaining Amp. LC Freq. Division Output Temp. Comp. RF-TCHO Reference oscillator Signal Conditioning Output 49 of 73

50 Reference Oscillator Transconductance amplifier + I(T) bias Resonant tank, LC f o (T) compensation module, C v+f (T) Process variation comp. module, C f (b p-1,,b 0 ) + _ -g m _ + + v _ R L L R C C f C v+f (v ctrl ) C v+f (v ctrl ) v ctrl (T) generation Resonant frequency correction, C f (b p-1,,b 0 ) I(T) generation On tester load board f ref b p-1,,b 0 generation Automatic frequency calibration macro 50 of 73

51 Resonant Tank, LC R L Due to the parasitic R L & R C present in a monolithic implementation: 1 R C ω LC L C R L (T) & R C (T) cause a temp. induced frequency drift: ω 1 Where: 2 CRL( T ) L ( T ) = ωo ω 2 o 1 CR ( T ) L C ωo = 1 LC CRL ( T ) L Temperature drift is highly linear and dominated by coil loss 2 51 of 73

52 Reference Oscillator V DD Complementary cross-coupled g m amplifier pmos tail to minimize flicker noise upconversion Cascode to minimize bias sensitivity R MR p x 11x x 11x R MR n 4nH +v out -v out v ctrl (T) I bias C v (v ctrl ) C v (v ctrl ) ½C f ½C f 52 of 73

53 Transconductance Amplifier + I(T) Bias Transconductance amplifier + I(T) bias Resonant tank, LC + _ -g m _ + + v _ R L L R C C f I(T) generation 53 of 73

54 Transconductance Amplifier i C (t) (ma) g m -amp injects current onto net capacitance t (ns) current 1 ω = ω Q n=2 Waveform is distorted voltage 2 n 2 n 1 2 o h i ( n) v C (t) (mv) Sustains oscillation by injecting energy (current) into the resonant tank Causes harmonic work imbalance which leads to frequency drift Frequency drift due to harmonic work imbalance function of normalized Fourier coefficients h i(n) of current waveform Note, as Q, drift due to harmonic work imbalance approaches 0 54 of 73

55 Frequency Drift Mechanisms PVTA frequency drift originates from Initial inaccuracy due to process variation (P) Harmonic work imbalance due to bias changes (V) TC due to coil loss (T) Aging due to package and common mode variation from hot carrier and tunneling effects (A) not discussed in this seminar (but terribly interesting) To achieve desired accuracy, must develop analog open-loop compensation circuitry 55 of 73

56 Temp. Comp., C v+f (T) Transconductance amplifier + I(T) bias Resonant tank, LC f o (T) compensation module, C v+f (T) + _ -g m _ + + v _ R L L R C C f C v+f (v ctrl ) C v+f (v ctrl ) v ctrl (T) generation I(T) generation 56 of 73

57 Temp. Comp., C v+f (v ctrl ) f o (T) compensation is programmable x-bit bank of AMOS varactors in parallel with fixed capacitance To one side of the resonant tank 2 x-1 C v 2 x-1 C f V DD b x-1 Control varactors with a temperaturedependent control voltage, v ctrl (T) creating a temperaturedependent capacitance, C v+f (T) v ctrl (T) 1C v V DD 1C f b 0 57 of 73

58 Temp. Comp., v ctrl (T) Create a temperaturedependent current, I(T), using a combination of temperature-dependent current generators SourceI(T) into a resistor with a known TC generating a temperaturedependent control voltage, v ctrl (T) I(T) v ctrl (T) R t-1 (T) I(T) To v ctrl (T) of C(v ctrl ) calibration module R 0 (T) C Include the ability to switch resistor types to allow v ctrl (T) to be finely tuned b t-1 b 0 58 of 73

59 Effects of C f+v (v ctrl ) & v ctrl (T) on f o (T) f o AMOS varactors enable coarse tuning Resistor TC bank enables fine tuning Tank is mostly variable capacitance, C v Tank is a combination of variable and fixed cap., C v +C f Linear negative f TC as predicted previously Tank is mostly fixed capacitance, C f T 59 of 73

60 Process Variation Comp., C f (b p-1,,b 0 ) Transconductance amplifier + I (T) bias Resonant tank, LC f o (T) compensation module, C v+f (T) Process variation comp. module, C f (b p-1,,b 0 ) + _ -g m _ + + v _ R L L R C C f C v+f (v ctrl ) C v+f (v ctrl ) v ctrl (T) generation Resonant frequency correction, C f (b p-1,,b 0 ) I(T) generation 60 of 73

61 Process Variation Comp., C f (b p-1,,b 0 ) To resonant tank 2 p-1 C trim bp-1 2 p-1 C trim bp-1 1C trim b 0 1C trim b 0 Parallel binary-weighted fixed capacitor banks Binary-weighted capacitor array adds or subtracts capacitance adjusting the oscillation frequency Simple concept; complicated details 61 of 73

62 Automatic Frequency Calibration Transconductance amplifier + I(T) bias Resonant tank, LC f o (T) compensation module, C v+f (T) Process variation comp. module, C f (b p-1,,b 0 ) + _ -g m _ + + v _ R L L R C C f C v+f (v ctrl ) C v+f (v ctrl ) v ctrl (T) generation Resonant frequency correction, C f (b p-1,,b 0 ) I(T) generation On tester load board f ref b p-1,,b 0 generation Automatic frequency calibration macro 62 of 73

63 Automatic Frequency Calibration A digital frequency locked loop (FLL) that runs counting races between a precision reference and the RF-TCHO CLK_REF MSB x-bit REF counter REF_MSB N 0 P Register P CLK_IN RESET TC x-bit CLK counter REF_RESET CLK_TC Up/down counter + State machine N 1 S RESET Bus CLK_RESET RF-TCHO EC 63 of 73

64 Reference Oscillator Transconductance amplifier + I 1 (T) bias Resonant tank, LC f o (T) compensation module, C v+f (T) Process variation comp. module, C f (b p-1,,b 0 ) + _ -g m _ + + v _ R L L R C C f C v+f (v ctrl ) C v+f (v ctrl ) v ctrl (T) generation Resonant frequency correction, C f (b p-1,,b 0 ) I(T) generation f ref b p-1,,b 0 generation Automatic frequency calibration macro 64 of 73

65 USB Implementation of RF-TCHO USB to RS-232 bridge controller for cables and thumb drives RF-TCHO replaced the XTAL + PLL with an all-si clock generator and reduced the clock module cost to pennies and size by over 1,000X 100kunits/month 400µm 450µm [McCorquodale, et al., JSSC, Feb. 2007] RF-TCHO is first commercial quartz replacement 0.18mm 2 in 0.35µm CMOS 65 of 73

66 Mobius 12MHz RF-TCHO : Temp. Comp. Frequency Error of RF-TCHO Frequency Error (ppm) Temperature ( C) Uncompensated Compensated 66 of 73

67 Mobius 12MHz RF-TCHO : Temp. Comp. Frequency Error of RF-TCHO Frequency Error (ppm) Temperature ( C) Nominal VDD VDD-10% VDD+10% 67 of 73

68 Performance Comparison Variable/Metric 24MHz XO 25MHz Si MEMS 24MHz Ceramic 12MHz RF-TCHO Total accuracy (ppm) ~±10 ~±25 ~±3200 ~±225 SSB phase noise (dbc/hz) -102/-124/ /-85/ /-129/ /-124/-141 RMS period jitter (ps) Power 68 of 73

69 Conclusions 69 of 73

70 Takeaways in clock generation Technical Conclusions Most reference clocks are at <100MHz ±>100ppm Power limits maximum frequency that touches PCB Device physics limits quartz scaling Most carrier synthesis refs. are at <20MHz ±<25ppm System observation takeaways In a PLL, the output performance is dictated largely by the output VCO, thus reference oscillator component Q becomes much less significant Frequency multiplication and division degrade and enhance phase noise and jitter substantially For clock jitter, far-from-carrier phase noise is more important than close-to-carrier phase noise 70 of 73

71 RF-TCHO Observations and Future Work RF-TCHO Takeaways RF-TCHO is essentially a stabilized free-running LCO which is equivalent to an LC-PLL Architecture guarantees low jitter and phase noise Close-to-carrier phase noise will still likely be higher as compared to high-q references, though not by much Frequency inaccuracy dominated by TC, not VDD or trimming inaccuracy Seek to develop compensation techniques to achieve <±50ppm inaccuracy Thought-provoking comments on RF-TCHO What will the start-up latency of a RF-TCHO be? So what? Can RF-TCHO be applied to RF? Can RF-TCHO be integrated to replace the channel-rate (as opposed to reference) clock generator? 71 of 73

72 Final Conclusions Quartz is likely to be replaced in the near term Several viable technologies now sampling commercially FBAR and RF-TCHO on market in volume production Si MEMS sampling Likely fragmentation of applications based on performance (accuracy + jitter / phase noise) Si MEMS: ±25ppm ±50ppm, but too high of phase noise for RF FBAR: Filters already in production Ceramic: ±500ppm ±5kppm still not a quartz replacement technology RF-TCHO : ~±50ppm ±500ppm, low jitter and sufficient accuracy for clocking and maybe more 72 of 73

73 The Race to Replace Quartz Thank you for your attention and enjoy the race! Questions are welcome 73 of 73