The Art and Science of VLSI Chip Design: Bridging the Analog(ue) and Digital Worlds Dr. Subhajit Sen DA-IICT, Gandhinagar The “Digital Divide” Analog(ue) or Physical Or Real World Digital Or Computer Or Virtual World Some statements About Analog vs. Digital • My 3G phone sounds better than your 2G phone • A “Digital” equipment/instrument is superior to analog equipment • A “Digital” camera gives better picture quality than a film camera • Analog electronic circuits are more complex: require more components than digital ones • Analog equipments consumes more power than “digital” ones • Analog electronic circuits are harder to design than digital ones • Nobody designs analog circuits/instruments any more The Analog World The Analog vs. Digital Signal • Infinite precision in amplitude and time axes • Absolute vs. relative “one” and “zero” in a digital signal • A digital signal is often “coded” (0 second, 2 Volt) 10111 10110 10101 01101 11… (1 second, -1 Volt) Stages of Progression from Analog to Digital • Sampling: – taking instantaneous time snapshots of changing phenomena • Quantization – Rounding off to integer or sub-integer amplitude intervals • Coding – Binary encoding, compression Information Rate while Sampling d v Rate = v d Minimum Sampling Rate d’ v Minimum Sampling Rate = v d’ Aliasing in sampled phenomena Aliasing in continuous-time signal • Leads to information loss (or rather misinformation) • Sampled sequence: 0.6, 0.9, 0.9, … Information rate or “Bandwidth” • Any phenomenon should be sampled at twice the bandwidth(or information rate) of the phenomenon – Sampling Rate = 2 X Bandwidth • Any information rate change higher than the bandwidth leads to “aliasing” and loss of information Interpolation • • • Needed for signal reconstruction Computation of predicted values Circuit needed to generate interpolated timeinstants Quantization • Inherently lossy • Needs accurate voltage reference • Typically combined with binary coding in circuits 1.00 V 0.75 0.5 0.25 0 -1.00 -0.75 -0.5 -0.25 V Coding • Binary encoding • Inherent part of analog-digital converters/codecs 1.00 0.75 0.5 0.25 0 -0.25 -0.5 -0.75 -1.00 0100 0011 0010 0001 0000 1100 1011 1010 1001 Coding: compression • More frequently used numbers or number sequences are encoded with short code words and vice-versa • Analogy with text message abbreviations: – “Wud b late 2nite, b4 12, 2 much wrk, lv” – “AFAIK he is frm Thiruvananthapuram” • Information loss Ideal Sampling Capacitor 00110 10111 11111 10010 A Sampling Circuit Clock Capacitor Track-and-Hold Circuit • Can use MOS transistor as switch Clock Capacitor Sample-and-Hold circuit • Bandwidth reduction necessary to avoid aliasing • Inaccuracies: gain-error, distortion error,noise Anti-alias Filter T& H T& H Capacitor 00110 10111 11111 10010 From Analog signal to Digital Codes • Anti-alias filtering • Sample-and-Hold • Quantization/Coding clock ADC Anti-alias SampleFilter/Amplifier And-Hold Reference voltage 10110 11011 1.. Encoder Analog Re-construction • Decoder • Reconstruction filter Reference voltage 10110 10111… Decoder Digital samples DAC Re-construction Filter Analog-Digital converter in VLSI chip VLSI Chip Analog or Physical Or Real World Reference Clock Reference Voltage ADC Anti-alias SampleFilter And-Hold Encoder Digital Or Computer OrDigital Virtual World 10110100… Or Computer Or Virtual World DAC Analog-Digital conversion “Raw” Analog Signal Low-pass (anti-alias) filtering Sampling and quantization Coding 00110 10111 11111 10010.. Clock Frequency Accuracy • Analog information is preserved ONLY if the clock frequency is accurate • Time period of an “on-chip” oscillator is governed by resistor/capacitor component values: R*C or L*C • Very wide manufacturing variations in frequency if R and C components are INSIDE the chip: upto 40% Reference Clock • Crystal Oscillator: Electronic tuning fork – Extremely accurate & stable clocks possible: fews seconds/year – Eric Vittoz – The Swiss connection Crystal Oscillator Reference Clock sub-division • Synchronization circuits or phase-locked-loops • Henri Bellescize • Arbitrary sub-division of clock is possible Analog re-construction • Interpolation involves digital computations • A PLL (phase-locked-loop) can give sub-divided clock Accuracy: Reference Voltage Sub-division • Very accurately matched components possible “on-chip” • Possible to improve precision using “calibration” 1.00 0.75 0.5 0.25 0 -0.25 -0.5 -0.75 -1.00 0100 0011 0010 0001 0000 1100 1011 1010 1001 Reference Voltage • Analog information is preserved ONLY if the reference voltage is accurate • Need an accurate “battery” on-chip • Use physical property of silicon: – Band-gap of silicon is 1.1 electron Volt at 0 degree Kelvin – Possible to design “on-chip” circuits to extract this voltage Conduction Band Bandgap=1.1 eV Valence Band Analog De-mystified (Complexity) An analog multiplier A digital floating-point multiplier Analog De-mystified (Complexity) Analog FM radio With antenna “Digital” Radio Analog De-mystified (Why Analog chip design is Hard) • Digital Scaling: supply voltage going down • Noise is a indispensable part of analog circuits • Signal to Noise ratio reduces 5V supply more headroom less 1V supply Old Analog Systems Modern VLSI Systems Analog De-mystified (Why Analog chip design is Hard) • • • • Digital Scaling: supply voltage going down Transistor non-linearity governed by voltage head-room Lesser the head-room more the distortion Signal to distortion ratio decreases 5V supply 1V supply Old Analog Systems Modern VLSI Systems Chip Design at DA-IICT • A sample-and-Hold circuit with a back-end “sigma-delta” ADC designed and fabricated • 180nm CMOS • 2 iterations: – first version had a bug – Second has been tested for functionality • Testing issues Sample-and-Hold Circuit designed at DA-IICT (180nm CMOS) Sample-and-Hold Circuit designed at DA-IICT (180nm CMOS) OP-AMP (A1) Clock_gen OP-AMP (A2) CLF=2 pF Cb=5 pF Cs=10 pF Sigma-Delta ADC designed at DA-IICT (180nm CMOS) Measurement waveforms on Chip Design at DA-IICT Conclusion: Is Digital Really Better Than Analog • Not always. Any signal transformation to digital domain always involves information loss due to aliasing, quantization and coding. If the signal source characteristics are not well known or power consumption is an issue use analog signal processing. • However, once analog signals are converted to digital domain they can be stored and transmitted without much degradation i.e. digital circuits are much more robust against noise and distortion when stored or transmitted. Thank You The Art and Science of VLSI Chip Design: Bridging the Analog and Digital Worlds Modern VLSI chip based computers become interesting only when interfaced with the analog real world of continuous-time signals using analog signal- processing circuits that convert from continuous-time signals to coded bits. This talk will explain the methods (algorithms) and circuit techniques of this conversion based upon the sound principles of information-theoretic principles of sampling,aliasing, quantization and coding. It will also be explained how a crystal-oscillato attached to the chip pins and the silicon “band-gap” voltage reference inside the chip play crucial role in preserving the accuracy of analog information inside a VLSI chip. The talk also explains the relative merits of analog and digital processing of real world signals. The talk then attempts to answer the critical question: given the increasing cost and complexity of chip design how much of digital processing is really necessary in electronic gadgets and instruments that sense, process, store and transmit real world analog signals Importance of Sinusoids • Fourier (spectrum) analysis • Analog Circuit distortion is detectable Over-sampling and decimation • To reduce aliasing we may sample at a rate higher than twice the signal bandwidth • Can reduce the sample rate at any later stage by discarding samples (decimation) – Need to reduce information rate using an antialias filter Coding: crytography “PvUmQ” “cAokadfofadsfxcliy” “I won one crore lottery” Computer • Key • Loss of key leads to loss of entire message