Ultimate Low Cost Analog BIST

1
Ultimate Low Cost Analog BIST
Universidade Federal do Rio Grande do Sul -
UFRGS Instituto de Informática Programa de
Pós-Graduação em Computação - PPGC Porto Alegre,
RS, Brazil

• Marcelo Negreiros, Luigi Carro, Altamiro A. Susin
• negreiro,carro,susin_at_inf.ufrgs.br

40th Design Automation Conference - DAC2003 June
02-06, 2003, Anaheim, California, USA
2
Introduction
SoC Testing
3
SoC testing - Analog BIST

• Desirable analog BIST features
• low analog area overhead digital overhead is
  preferred
• reuse system resources minimal system overhead
• few analog measurement points
• low performance degradation
• low cost.

4
Analog testing (1)

• HIGH COST

Analog circuit
5
Analog testing (2)

• VARIABLE LOAD
• TEST SERIALIZATION

Analog circuit
6
Analog testing (3)
1-bit ADC

• LOWEST COST
• PARALLELISM

1-bit ADC
1-bit DAC
1-bit ADC
Analog circuit
7
Analog BIST approach based on a low cost sampler
and signal generator
AD
Analog Core
DSP
Processor
DA
Test Core
Mem
Digital
Digital
Core B
Core A
SoC

• multiple test points can be analyzed
• analog switches/muxes are not required
• the analog circuit observes a constant load.

8
Presentation topics

• Power Spectrum Density (PSD) based analog test
• Test example
• Analysis
• Conclusions

9
PSD-based analog testsystem identification

• Transfer function identification based on noise
  (all frequency band is excited at the same time)

frequency domain
10
PSD-based analog testwhat is required?
PSD

• Reference PSD
• (fault-free circuit)

frequency
PSD

• CUT PSD
• (faulty circuit?)

frequency
11
PSD-based analog testDistance measurement
PSD
frequency
12
PSD-based analog testDistance threshold
PSD
frequency
13
PSD-based analog testDistance evaluation
Distance
dt
Distance threshold
0
CUT PSD deviation from reference PSD ()
14
Ultimate low cost bist requirements1-bit DAC
and 1-bit ADC
Analog circuit
1-bit DAC
1-bit ADC
1-bit ADC
Test core
15
Single-bit quantization of noise

• If the input signal is a normal stationary
  process (zero mean)

The autocorrelation is transferred through the
non-linearity, by a scaling factor.
16
Signal generator output PSD

• multi-bit and 1-bit gaussian noise
• Matlab simulation

17
ADC input PSD comparison

• Linear filter
• Quality factor (Q) changes -90,-50,0,50,90
  

90
90
0
0

• multi-bit data

• one-bit data

• power normalization

18
Distance obtained from previous PSD comparison

• Linear filter
• Quality factor (Q) changes -90,-50,0,50,90
  

90
0
90
0

• multi-bit data

• one-bit data

19
Test example
Continuous time state variable filter ITC97
analog and mixed-signal benchmarks
from http//www.coe.uncc.edu/cestroud/analogbc/mi
xtest.html
20
Testing a state variable filter
filter
generators HP33120A
data acquisition AD 2181 ezkit
comparators
21
Test details

• 9 passive components R1-R7,C1,C2
• Faults inserted
• catastrophic open/short (18)
• large parametric 50 (18)
• small parametric 20 (18)
• Evaluated test performance for
• 1) 12800 samples
• 2) 51200 samples
• 3) 16-bit AD, 8-bit DA and 12800 samples

22
Test results faults not detected
1) 12800 samples
Faults
open/short
0/18
50
5/18
20
18/18
23
Analysis (1)

• Test comparison 1-bit x multi-bit processing
• 1-bit 23 out of 36 parametric faults not
  detected
• multi-bit 9 out of 36 parametric faults not
  detected
• Both have detected all open/short faults
• Increasing test time (4x)
• 1-bit 7 out of 36 parametric faults not detected

24
Analysis (2) Comparator Offset
Small influence on the distance for usual values
of offset.
25
Analysis (3)

• smallest sampler and signal generator (minimal
  additional analog area)
• minimum performance degradation
• (constant load, no switches or muxes)
• enables the observation of several test points
• allows reuse of resources in a SoC environment.
• low cost

26
Conclusions

• A PSD-based test for analog circuits using 1-bit
  DAC and ADC was presented
• Small degradation of the test when comparing to
  the multi-bit case
• Trade-off analog area x test time
• Lower test time at the system level because of
  parallelism
• Further work non-linear systems.