CN1018091B - Noise component analysis and measure system - Google Patents

Abstract

The present invention relates to a semiconductor device low-frequency noise element analysis method and a corresponding test system, particularly to an automatic test system with a microcomputer for treatment. The present invention uses low-frequency noise for predicting the long term use reliability of semiconductor devices, so the present invention provides a novel rapid reliable method without damage for the reliability screening of the semiconductor devices. Thus, the present invention realizes accurate quantitative analysis on low-frequency noise elements for the first time, and has initiated contribution to semiconductor device noise theory establishment, device inherent defect analysis and the reliability screening.

Description

Noise component analysis and measure system

The present invention relates to a kind of analysis test method and corresponding testing device to transistor low-frequency noise composition, particularly, there is the automatic testing equipment of MICROCOMPUTER PROCESSING in generation.

To transistorized low-frequency noise the composition size and the corner frequency analysis of (comprise white noise, 1/f noise, reach the g-r noise), be to understand defective in device surface and the body, the Noise Mechanism of research device reaches unreliable transistorized harmless, a kind of important method of rapid screening.This invention is not seen the document pertinent to the point of pair quantitative test of each composition of low-frequency noise to the world patents index retrieval.The retrieval engineering index, wherein have only one piece more close, its exercise question:

" (AlGa) AS-GaAS2-D electronic gas pipe is carried out the deep level analysis by means of the low-frequency noise measurement.”

(original text: " and Deep-level analysis in(AlGa) AS-GaAS2-d electron gas devices by means of law freguououey mise measuremonfs ").

This article is to observe from the noise spectrum curve to the method that the g-r noise component analysis in (AlGa) AS-GaAs element low-frequency noise adopts, and is discerned by the projection that the g-r noise causes, not only observe not obvious, and determined corner frequency f ₀Also very difficult, therefore, can't obtain g-r noise, 1/f noise, and the amplitude of white noise.About corresponding testing device, by retrieval, in the prior art as HP4470 transistor, the field effect transistor SONOAN of the U.S., it can not finish transistor low-frequency noise composition analysis, it is to the test of noise spectrum, can only be with manually changing frequency, pointwise is finished, and not only the test duration is long, and very inconvenient.Its test low-limit frequency is 10HZ, to the amplifier gain K(f of noise measuring set) proofread and correct and all adopt AGC method, low precision.This invention can be finished the quantitative test of low-frequency noise composition automatically, frequency range is at 0.25HZ-100KHZ, every frequency range is established and is differentiated point 400 points, average time N minimum is elected as 512 times, and 1024 times, and the amplitude versus frequency characte K(f of pair amplifier) measure automatically, secondly noise spectrum is measured, by the IEEE-488 interface with statistical average method, deliver in the computing machine, finish Si(f) calculating and noise component analysis.Yet precision height, easy to use.Therefore, analytical approach and the corresponding testing device from noise is domestic initiation.

Main task of the present invention: be that a kind of transistor low-frequency noise analysis of components method will be provided, and set up the proving installation that there is MICROCOMPUTER PROCESSING in corresponding quantitative generation.Promptly come prediction transistor device long term reliability, thereby be the reliability screening of transistor device, accurate, quick, reliable new tool is provided with low-frequency noise.

Task of the present invention realizes with following method:

(ⅰ) set up transistor low-frequency noise analysis of components method:

According to asking several noise components in the device equivalence input current noise:

$S_i\mathrm{(f)\; =\; A\; +}\frac{B}{f}+\underset{\mathrm{i\; =\; 1}}{\overset{N}{\mathrm{\&Sigma;}}}\frac{C_i{\mathrm{f/\; <figure-callout\; id="1"\; label="f\; oi"\; filenames="88100045\_DRIMG10.png,88100045\_DRIMG11.png"\; state="\{\{state\}\}">f\; </figure-callout>}}_{\mathrm{oi}}}{{\mathrm{1+(f\; /f}}_{\mathrm{oi}}{)}^2}$

(1) the amplitude B of 1/f noise, i.e. expression-form B/f;

(2) the amplitude C of g-r noise _i, corner frequency f _Oi, i.e. expression formula

$\underset{\mathrm{i\; =\; 1}}{\overset{N}{\mathrm{\&Sigma;}}}\frac{C_i{\mathrm{f\; /<figure-callout\; id="1"\; label="f\; oi"\; filenames="88100045\_DRIMG10.png,88100045\_DRIMG11.png"\; state="\{\{state\}\}">f\; </figure-callout>}}_{\mathrm{o\; i}}}{{\mathrm{1\; +\; (f\; /\; f}}_{\mathrm{oi}}{)}^2}$

(N is different energy level defective compositions for transistor inside has).

(3) amplitude of white noise is A.

(ⅱ) for using the custom-designed noise power spectrum proving installation of said method:

For the low-frequency noise composition of analyzing crystal pipe, the amplitude versus frequency characte K(f of necessary pair amplifier) and equivalent input noise current power spectrum S _i(f) measure.As shown in Figure 26:

By input circuit (comprising tested transistor), measuring amplifier, CF-920 analyser, IEEE-488 interface circuit, SUPer micro computer, printing show and positive 9 volts of power supplys are formed.

At first, produce the scanning sine wave by the CF-920 analyser and send into amplifier, obtain amplifier amplitude versus frequency characte K(f by CF-920 again), secondly, tested transistor noise is sent into the CF-920 analyser after measuring amplifier amplifies, be noise power spectrum S _i(f) measure,, in the 0.25HZ-100KHZ scope, divide three band frequency interval measures: (a) 0.25HZ-95HZ in order to guarantee enough little frequency interval; (b) 100HZ-4950HZ; (c) 5000HZ-100KHZ.All test noise data and K(f) totally 315 points, through the IEEE-488 interface circuit, send into computing machine (SUPer) and do corresponding calculated, obtain fs _i(f), and then pass through noise component analysis software, obtain the various compositions of noise.As accompanying drawing 13 is whole testing software FB(flow block) and subroutine such as accompanying drawing 14-25.

(1) finishes S _iThe technical indicator of measurement (f):

A) frequency range: 0.25HZ-100KHZ

B) survey frequency is counted: 315 points.

C) measuring accuracy:＜4%.

D) Measuring Time: 1 hour.

E) can measure the minimum current noise spectral density:

2.95×10 ^-13A/ $\sqrt{\mathrm{HZ}}$ （I _b＝20μA;β＝62;f＝1HZ）。

(2) finish by S _i(f) separate the calculating of three kinds of noise components in, print job, must finish following work by computing machine for this reason:

A) to S _i(f) carry out conversion fS _i(f) after, print curve, find out the frequency f of peak dot correspondence thus _OiAs accompanying drawing 10(a, b).

B) according to S _i(f) the known f of formula _oAfter, promptly optional white noise accounts for the S that Main Ingredients and Appearance and 1/f noise account for two Frequency points of Main Ingredients and Appearance _i(f) value is found the solution simultaneous equations, obtains A, B, C _i

C) according to A, the B, the C that obtain _i, f _oObtain fS _i(f) curve, and with the fS that measures _i(f) value is simultaneously displayed on the screen, when the two is inconsistent, readjusts f _Oi, up to the two error of fitting less than 5%.

D) print demonstration A, B, C _iAnd f _OiValue.

Accompanying drawing 1 and accompanying drawing 2-9 have provided the software flow block diagram of whole analytical approach.

The source resistance R _SSelect 100K Ω, so I ² _nR ² _n＞＞4KTR _S, measured is current noise.When surveying different pipe, all with regulating I _bBe the value of 20 μ A(or other regulation).Thereby the defect size that can compare each device.Finish the measurement of amplifier frequency response with the CF-920 analyser, the scanning sine wave of being sent by the CF-920 analyser is sent the input end of CF-920 analyser again back to after amplifier amplifies.Automatically obtain K(f) amplitude.

The present invention has realized the quantitative test to transistor low-frequency noise composition first, and to setting up transistor noise theory, the analysis of device latent defect, reliability screening is all significant.This characteristic feature of an invention:

1. three kinds of compositions in can analyzing crystal pipe low-frequency noise: white noise, 1/f noise, and g-r noise.

2. can finish low-frequency noise composition in the 0.25HZ-100KHZ scope automatic test, analyze, have Measuring Time is short, accuracy is high (directional error and stochastic error are little), measuring process full-automation.

3. measure and analysis result, can provide by demonstration, printing type.

4. survey frequency low frequency of the present invention can be at 0.25HZ, and every frequency range is differentiated maximum 400 points of point.

5. the present invention adopts the amplitude versus frequency characte K(f of pair amplifier) measure automatically, use computing method then, i.e. S _i(f)=S _o(f)/K ²R ² _SObtain, so the accuracy height.

6. the present invention adopts the noise spectrum statistical average method 512,1024 times, so stochastic error is little.

Yet the present invention to transistorized quantitative test, research, has made the contribution of digging property for setting up transistor low-frequency noise theory for both at home and abroad.

Concrete grammar of the present invention and structure are provided by following examples and accompanying drawing:

Fig. 1: data processing software block diagram;

Fig. 2: be the subroutine 1 of Fig. 1;

Fig. 3: be the subroutine 2,3,4,5,6 of Fig. 1;

Fig. 4: be the subroutine 11 of Fig. 1;

Fig. 5: be the subroutine 12 of Fig. 1;

Fig. 6: be the subroutine 21 of Fig. 1;

Fig. 7: be the subroutine 22 of Fig. 1;

Fig. 8: be the subroutine 221 of Fig. 1;

Fig. 9: be the subroutine 222 of Fig. 1;

Figure 10: be the frequency characteristic figure of noise spectrum;

Figure 11: be transistor noise E _n-I _nEquivalent circuit diagram;

Figure 12 (a, b): be CF-920 analyser test noise random character figure;

Figure 13: be whole testing software process flow diagram;

Figure 14: be the subroutine 1,2 of Figure 13;

Figure 15: be the subroutine 3 of Figure 13;

Figure 16: for the subroutine 4,5 of Figure 13 ... 9;

Figure 17: for the subroutine 11,12 of Figure 13 ... 16; 21,22 ... 27;

Figure 18: be the subroutine 111 of Figure 13;

Figure 19: be the subroutine 113 of Figure 13;

Figure 20: be the subroutine 115 of Figure 13;

Figure 21: be the subroutine 112 of Figure 13;

Figure 22: be the subroutine 114 of Figure 13;

Figure 23: be the subroutine 116 of Figure 13;

Figure 24: be the subroutine 1111 of Figure 13;

Figure 25: be the subroutine 1112 of Figure 13;

Figure 26: be noise power spectrum proving installation block diagram.

Below in conjunction with accompanying drawing, concrete grammar proposed by the invention and proving installation are described.

Accompanying drawing 11 is the transistor noise equivalent electrical circuit, E _n-I _nNoise model, R _SBe resistance, E _nBe the effective value of equivalent input voltage noise, I _nBe equivalent input current noise effective value.Corresponding power spectrum density (is the E of unit bandwidth ² _nWith I ² _n) be respectively:

$S_i\mathrm{(f)\; =\; 2q\; I}{}_b+\frac{{\mathrm{2qf}}_L{I}^r{}_b}{f}+\underset{\mathrm{i\; =\; 1}}{\overset{N}{\mathrm{\&Sigma;}}}\frac{{\mathrm{KI}}^2{}_b{\mathrm{/\; <figure-callout\; id="1"\; label="f\; oi"\; filenames="88100045\_DRIMG10.png,88100045\_DRIMG11.png"\; state="\{\{state\}\}">f\; </figure-callout>}}_{\mathrm{o\; i}}}{{\mathrm{1\; +\; (f\; /\; f}}_{\mathrm{oi}}{)}^2}$

In the following formula preceding paragraph be white noise (comprise thermonoise, and Johnson noise), last two be corresponding 1/f noise, with the g-r noise, if device has several g-r noises, several f are just arranged respectively then _o, purpose will be obtained the size of the 1/f and the g-r noise of device, thereby obtain the amount of reflection defect level, i.e. f _LAnd K value.

(1) transistor low-frequency noise analysis of components method:

From the noise power spectrum S that has recorded _i(f) promptly can determine corresponding white noise, 1/f noise, and the size of g-r noise intensity and defective.In the prior art, be to adopt the observation noise spectrum to discern in the projection of a certain Frequency point, as (a) in the accompanying drawing 10, for less projection, even do not have projection, and the g-r noise is arranged equally, therefore adopt S in method of the present invention _i(f) do conversion:

${\mathrm{fS}}_i\mathrm{(f)\; =\; A\; f\; +\; B+}\underset{\mathrm{i\; =\; 1}}{\overset{N}{\mathrm{\&Sigma;}}}\frac{C_i{\mathrm{f\; /\; <figure-callout\; id="1"\; label="f\; oi"\; filenames="88100045\_DRIMG10.png,88100045\_DRIMG11.png"\; state="\{\{state\}\}">f\; </figure-callout>}}_{\mathrm{o\; i}}}{{\mathrm{1\; +\; (f\; /\; f}}_{\mathrm{oi}}{)}^2}$

As seen the g-r noise obviously is reflected in the projection of curve.When no g-r noise, fS _i(f)=Af+B is straight line I.FS when the g-r noise _i(f) curve has the peak to rise, so the present invention adopts Computer Processing, shows fS by computing machine _i(f) curve is at first found out fS _i(f) peak dot of curve (as (b) in the accompanying drawing 10, visible g-r noise obviously is reflected in curved convex).If any several f _o, several g-r noises are promptly arranged.And then obtain A, B, C according to measured data _iValue.The A that calculates, B, C _i, f _OiValue is obtained fS _i(f) curve, and with the actual measurement fS _i(f) be presented on the screen,, readjust f if the two does not overlap _OiValue is up to the two error of fitting＜5%.Thereby finished the low-frequency noise analysis of components.In order further to improve the analysis result of accuracy, key is the precision to the noise spectrum measurement, the noise spectrum that this test device system adopts CF-920 to record, adopt mean value 512 times, 1024 times, adopt the microcomputer automatic testing equipment, as accompanying drawing 1 is the data processing software block diagram, and subroutine, as accompanying drawing 2-9.The result who finishes makes relative error less than 4%, although it seems, the time is more longer, in the feeble signal system, and compares in the prior art, and the test duration is very short, particularly exchanges the raising of accuracy for.

(2) for using the proving installation of the custom-designed transistor low-frequency noise of said method power spectrum.

As shown in Figure 26, by input circuit (1), it is connected to measuring amplifier (2), CF-920 analyser (3), TEEE-488 interface circuit (4), reaches micro computer SUPer(5) and output terminal printer (6) and positive 9 volts of power supplys composition transistor noise power spectrum proving installation.

Accompanying drawing 26 is equivalent input noise current power spectrum S _i(f) block diagram of proving installation, this proving installation has not only guaranteed the hardware unit realization, and will work out corresponding testing software, guarantees that proving installation has try one's best high measuring accuracy and short test duration, the technical indicator of its test for this reason:

1. frequency range: 0.25HZ-100HZ;

2. measure dot number: 315 points;

3. measuring accuracy:＜4%;

4. Measuring Time: 1 hour;

5. can measure the minimum current noise density: 2.95 * 10 ^-13A √ HZ.

Test condition: I _b=20 μ A

β＝62

f＝1KHZ

Through CF-920 analyser, noise power spectrum S _o(f) measure, frequency be divided into three frequency range intervals:

① 0.25HZ-95HZ

② 100HZ-4950HZ

③ 5000HZ-100KHZ

All test datas are totally 315 points, through the IEEE-488 interface circuit, deliver to computing machine (SUPer) and do corresponding calculated and obtain fS _i(f), then by the noise component analysis element, obtaining the various compositions of noise, is whole testing software FB(flow block) as accompanying drawing 13, and its accompanying drawing 14-25 is the subroutine of testing software FB(flow block).This device is developed in order to study device reliability and low-frequency noise relation, if this device is used for the factory scene, do not need to measure whole noise spectrum, and only need survey several g-r characteristics of noise points (as 1HZ, 20HZ, 500HZ ...) then surveying instrument and Measuring Time all can shorten and simplify greatly.

Yet the present invention is the quantitative test that realizes the low-frequency noise composition first, and to setting up transistor noise theory, transistor surface and inherent vice research, reliability screening waits all significant.

Claims (6)

1, a kind of transistor noise analysis of components method of testing is characterized in that:

A) with the quantitative measurement of spectrum analyzer tested transistorized amplifier output noise power spectrum S is housed _o(f) and amplifier voltage gain K (f),

B) use the IEEE-488 interface circuit, the S that spectrum analyzer is obtained _o(f) and K (f) data send into microcomputer;

C) according to formula S _i(f)=(S _O(f))/(K (f _O) ²R ² ₅) the transistorized current noise power spectrum S of calculating _i(f);

D) form according to the transistor noise composition:

$S_i\mathrm{(f)\; =\; A\; +}\frac{B}{f}+\underset{\mathrm{i\; =\; 1}}{\overset{N}{\mathrm{\&Sigma;}}}\frac{C_i{\mathrm{f/\; f}}_{\mathrm{oi}}}{{\mathrm{1+(f\; /f}}_{\mathrm{oi}}{)}^2}$

To S _i(f) do conversion:

${\mathrm{fS}}_i\mathrm{(f\; )\; =\; Af\; +\; B\; +}\underset{\mathrm{i\; =\; 1}}{\overset{N}{\mathrm{\&Sigma;}}}\frac{C_i{\mathrm{f\; /f}}_{\mathrm{o\; i}}}{{\mathrm{1\; +\; (f\; /\; f}}_{\mathrm{oi}}{)}^2}$

FS by screen display _i(f) curve is determined the peak dot frequency f _Oi, conduct the S at frequency low side and high-end two Frequency point places again a by-election _i(f) value is by S _i(f) expression formula goes out A, B, C by Microcomputer Calculation _iValue;

E) according to A, the B, the C that calculate _iAnd f _OiValue is obtained fS _i(f) with the fS that surveys _i(f) curve is simultaneously displayed on the screen, by adjusting f _OiMethod, guarantee curve fitting error＜5%.

F) to the physical significance of the symbol in the right 1:

(1) f ₀(f)-amplifier output noise power spectrum (unit: V ²/ HZ).

(2) K (f)-amplifier voltage gain (dimensionless).

(3) S _i(f)-transistor current noise power spectrum (unit: A ²/ HZ).

(4) R _S-source resistance (unit: Ω).

(5) f-frequency (unit: HZ).

(6) S _Oi-s-r noise corner frequency (unit: HZ).

(7) A-white noise amplitude (unit: A ²/ HZ).

(8) B-1/f noise amplitude (unit: A ²).

(9) Ci-s-r noise amplitude (unit: A ²).

(10) N-s-r noise source number.

2, by the described transistor noise analysis of components of claim 1 method of testing, it is characterized in that: the frequency range 0.25HZ-100KHZ of tested noise power spectrum.

3, by claim 1 or 2 described transistor noise analysis of components method of testings, it is characterized in that: noise power spectrum and the quantitative measurment of amplifier voltage gain are divided into three band frequency scopes and carry out:

（1）0.25HZ-95HZ;

（2）100HZ-4950HZ;

（3）5000HZ-100KHZ。

4, by the described transistor noise analysis of components of claim 1 method of testing, it is characterized in that: transistor noise power spectrum quantitative test adopts the noise spectrum statistical average method, and its average time N elects 512,1,024 two kinds as.

5, a kind of is to use the described transistor noise analysis of components of claim 1 method of testing, and custom-designed transistor noise proving installation is characterized in that:

This proving installation is made up of input circuit (1), measuring amplifier (2), spectrum analyzer (3), IEEE-488 interface circuit (4), microcomputer (5), printer (6).

6, by the described transistor noise proving installation of claim 5, it is characterized in that: except by computer automatic analysis, outside the transmission, to result of calculation, this device also has automatic printing, display device.

Images