IC MACROMODELS FROM ON-THE-FLY TRANSIENT RESPONSES

Size: px

Start display at page:

Download "IC MACROMODELS FROM ON-THE-FLY TRANSIENT RESPONSES"

Esther Ball
6 years ago
Views:

1 IBIS DATE, Mar. 1, 26 MACROMODELS FROM ON-THE-FLY TRANSIENT RESPONSES F.G.Canavero, I.A.Maio, I.S.Stievano Dipartimento di Elettronica, Politecnico di Torino, Italy

2 IBIS DATE, Mar. 1, 26 Review of model generation (i) e.g., output buffer (single-ended) CORE vin In any approach, 2-piece model representation = w H (t) i H (v,d/dt) + w L (t)i L (v,d/t) i H,L : submodels accounting for buffer fixed logic H and L state w H,L : weighting signals for state switchings suitable modifications for handling power/ground pins and different device technologies 2

3 IBIS DATE, Mar. 1, 26 Review of model generation (ii) Classification = w H (t) i H (v,d/dt) + w L (t) i L (v,d/dt) Equivalent circuits Input Output Buffer Information Specification, IBIS Nonlinear parametric relations Macromodeling via Parametric Identification of Logic Gates, Mπlog parametric relations approximate any nonlinear dynamic system theory/tools from system identification improved accuracy for recent devices IBIS compliant (ver. 4.1) F i(k) = F(i(k-1),,v(k), v(k-1), ) 3

4 IBIS DATE, Mar. 1, 26 Mπlog modeling process (i) Parameter values obtained from suitable responses device ports by matching reference and model responses = w H (t) i H (v,d/dt) + w L (t) i L (v,d/dt) for i H,L : transient responses to suitable voltage fixed state + standard algorithms CORE H,L + - for w H,L : transient responses while the port is connected to reference loads and performs state transitions (e.g, 5 Ω resistor) + linear inversion of model equation 4

5 IBIS DATE, Mar. 1, 26 Mπlog modeling process (ii) Estimation of submodels i H,L 4 stat. H e.g., for i H CORE H in High output state ma + - i ma stat. L V t ns t ns 1 estimation algorithm (e.g., [1]) [1] M.T.Hagan et Al. Training feedforward networks with the marquardt algorithm, IEEE Trans. on NNs, Nov

6 IBIS DATE, Mar. 1, 26 Mπlog modeling process (iii) Computation of weighting signals w H,L CORE driven to perform complete state transitions lumped load i ma 4 5Ω 1Ω Ω+VDD w H,L obtained by linear inversion of model equation = w H (t) i H (v,d/dt) + w L (t) i L (v,d/dt) by using 1,2, set of responses (solution of a standard LS problem) 6

7 IBIS DATE, Mar. 1, 26 Mπlog modeling process (iv) logic state must be controlled! CORE + - different sources and load conditions needed! CRITAL for model generation of complex s How to obtain waveforms useful for model estimation from device operating in normal condition? 7

8 IBIS DATE, Mar. 1, 26 Mπlog models from measurements on-the-fly (i) Typical structure 4 real board i ma and have pieces of information on both state transitions and device fixed state (limited region explored) 8

9 IBIS DATE, Mar. 1, 26 Mπlog models from measurements on-the-fly (ii) Typical structure + perturbing element 4 2 i ma Simplest (passive) perturbing element: stub (time delay > 1/3 1 bit time) and explore a more wide region of the solution space waveforms for both estimation of submodels and of weighting signals 9

10 IBIS DATE, Mar. 1, 26 Mπlog modeling process on-the-fly (i) Estimation of submodels i H,L = w H (t) i H (v,d/dt) + w L (t) i e.g, for i H + estimation algorithm (e.g., [1]) i ma t ns t ns 1

11 IBIS DATE, Mar. 1, 26 Mπlog modeling process on-the-fly (ii) Computation of weighting signals w H,L = w H (t) i H (v,d/dt) + w L (t) i L (v,d/dt) 1.5 e.g, for w L (t) = (1-w H (t)) i ma w H (t) = (-i L (v,d/dt)) / (i H (v,d/dt)-i L (v,d/dt)) t ns 11

12 IBIS DATE, Mar. 1, 26 Summary of Mπlog model generation (1) Device is conveniently stimulated and reaction is recorded + - Device mounted directly on the board and responses are recorded (2) Port responses feed an algorithm for the computation of model parameters (3) Model equation F is implemented in SPE or in metalanguages like VHDL-AMS i(k) = F(i(k-1),,v(k),v(k-1), ) F.cir F.vhd 12

13 IBIS DATE, Mar. 1, 26 Modeling example Modeled device 8-bit bus transceiver, SN74ALVCH16973, VDD=1.8V, bit time: 6ns Model 1: estimation using noiseless signals responses from SPE simulation Model 2: estimation in a noisy environment responses from SPE simulation + superimposed noise feasibility for estimation from measured data (i H,L are sums of 2 5 sigmoidal functions, dynamic order 2) 13

14 IBIS DATE, Mar. 1, 26 Model 1 validation Model built from noiseless signals 2 1 e.g., for the estimation of i H t 4 Static characteristics 1.5 Transient responses i ma reference (Hspice), macromodel (Spice implementation) 1.5 Load: ideal line (Z=5Ω, Td=1ns) + 1 Ω // 5pF tns 14

15 IBIS DATE, Mar. 1, 26 Model 2 validation Model built from noisy signals 2 1 e.g., for the estimation of i H σ = 4mV t i ma 4 2 Static characteristics Transient responses -2.5 Load: ideal line (Z=5Ω, Td=1ns) Ω // 5pF tns reference (Hspice), macromodel (Spice implementation) 15

16 IBIS DATE, Mar. 1, 26 Conclusions Mπlog model generation from measurements carried out on-the-fly allows the modeling of complex s from transient responses recorded during normal activity does not need any control of logic state minimizes experimental cost (no dedicated test fixtures) enables accurate estimation of device static characteristics from transient measurements 16

BLACK-BOX MODELING OF DIGITAL DEVICES

BLACK-BOX MODELING OF DIGITAL DEVICES I. S. Stievanol, I. A. Maiol and F. G. Canaveral Abstract The design for signal integrity and electromagnetic compatibility of modern fast electronic circuits heavily