RTN Induced Frequency Shift Measurements Using a Ring Oscillator Based Circuit Qianying Tang 1, Xiaofei Wang 1, John Keane 2, and Chris H. Kim 1 1 University of Minnesota, Minneapolis, MN 2 Intel Corporation, Hillsboro, OR Symposia on VLSI Technology and Circuits
Background Outline Proposed ROSC Based RTN Characterization Circuit 65nm Test Chip RTN Data Transistor V t Shift Estimation Summary Slide 1
Random Telegraph Noise (RTN) Basics Random trapping and de-trapping of carriers from channel Manifests as a fluctuation in V t resembling a random telegraph signal Slide 2
RTN Characterization Techniques Probing Individual transistor SRAM VDD Circuit based methods Logic circuit or ring oscillator Schematic WL WL D Q BL BLB Parameter of interest DC current SRAM V min Frequency shift Pros 1. Simple 2. High resolution 1. Realistic RTN impact on SRAM V min 2. Short test time 3. Small test area 1. Realistic RTN impact on circuit frequency 2. Short test time 3. Small test area Cons 1. Long test time 2. Large test area 3. Limited insight on circuit level 1. Limited resolution 2. Rare occurrences 3. Averaging effect 1. Limited resolution 2. Long meas. time 3. Averaging effect Slide 3
Simple ROSC Based Meas. Circuit Measure the divided ROSC frequency using offchip instrument A simple & popular structure for characterizing process variation and reliability issues No reported RTN data using this technique Single ended sensing, common mode noise, long measurement time poor resolution Slide 4
Beat Frequency Detection Scheme C A B T. Kim, et al., JSSC, 2008 D flip-flop captures the beat frequency of two free running ROSCs Benefits of beat frequency detection scheme Achieves <0.01% frequency measurement resolution at sub-microsecond sampling times Insensitive to common mode noise such as V, T drifts Fully-digital scan-based interface Slide 5
RTN Measurements Using BFD Scheme RTN Slide 6
RTN Measurements Using BFD Scheme RTN Slide 7
RTN Measurements Using BFD Scheme RTN Slide 8
RTN Measurements Using BFD Scheme RTN Small frequency shifts induced by RTN amplified Sub-ps resolution + sub-µs measurement time Slide 9
Comparison with Prior Art Counter Counter Counter [2] K. Ito, et al., IRPS, 2011 Slide 10
ROSC Based Test Chip Diagram Row Select DUT ROSC DUT ROSC DUT ROSC Column Select DUT ROSC DUT ROSC DUT ROSC row<n> col<m> DUT ROSC DUT ROSC DUT ROSC N:1 Mux Ref. ROSC A B A B 8bit Counter DFF D Beat Frequency Detection Unit Q C Reset Edge Detector Parallel/serial shift register Scan out An 8 10 DUT ROSC array, a shared reference ROSC and a beat frequency detection block Measure beat frequency between DUT ROSC and shared reference ROSC Slide 11
65nm Test Chip Die Photo Slide 12
ROSC Frequency Shift Measurements Single trap induced ROSC frequency shift is roughly 0.4% at 0.8V Capture/emission time constants range from 1 ms to 100 s of milliseconds Slide 13
Single and Multi Trap RTN Time Lag Plot (TLP) used to identify the number of RTN traps [3] [3] T. Nagumo, et al., IEDM, 2009 Slide 14
Frequency Shift versus Voltage f/f (%) Peak-to-peak f/f (%) Frequency shift due to RTN trap decreases at higher voltages Frequency shift sensitivity decreases Fermi level shifts with supply voltage Slide 15
Time Constant Distribution τ c =3.68 ms τ e =14.0 ms τ c =24.9 ms τ e =8.12 ms τ c =26.7 ms τ e =8.70 ms Capture and emission time fit exponential distribution under AC condition Pr = A exp(-t/τ) Slide 16
Time Constant versus Voltage τ c and τ e show opposite dependences τ c has negative dependence on voltage for NMOS and positive dependence for PMOS [3] [3] T. Nagumo, et al., IEDM, 2009 Slide 17
Power Spectral Density of RTN PSD of RTN induced frequency shift suggests a Lorentzian spectrum Slide 18
Number of Traps Distribution Most ROSCs do not show any RTN behavior No more than two RTN inducing traps observed Slide 19
Frequency to V t Mapping Frequency shift is proportional to V t shift V t shift due to single RTN trap is 1.9% for NMOS and 1.6% for PMOS Slide 20
V t Shift versus Supply Voltage Freq. shift reduces at higher supply voltages However, trap induced V t shift has a weak dependency on supply voltage Slide 21
Summary Beat frequency detection system utilized for high precision RTN measurement with short measurement time ROSC frequency shift due to RTN measured for the first time from a 65nm test chip Single trap RTN causes 0.4% frequency shift at 0.8V Wide range of time constants measured Frequency shift and time constant measured at different supply voltages Measured ROSC frequency data can be readily mapped to transistor V t shift Slide 22