JP2001215255A - Estimating method and estimating device for software error resistance of semiconductor device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a simple method for estimating the rate of software errors of a semiconductor device relative to cosmic ray neutrons. SOLUTION: The spectrum of cosmic ray neutrons is divided. The software errors are weighted by flux in each of divisional energy bands generated when the semiconductor device is irradiated with neutrons of typical energy. The total cross-section of the weighted software errors is used to forecast the rate of software errors undergone by the semiconductor device in an actual environment.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and apparatus for evaluating soft error resistance of semiconductor devices caused by cosmic ray neutrons.

[0002]

2. Description of the Related Art The occurrence rate _SQC of a semiconductor device soft error (SE, inversion of stored contents due to an unexpected cause) in quality evaluation is FIT (Failure in Time; 10 ^9).
When an error occurs once a time, it is 1 FIT. ), And the permissible limit for mass production is usually about 1000 FIT. Errors caused by radiation are SEU (Single Event
Upset, a soft error corresponding to one incident particle). The soft error rate (SER) is the number of errors that occur per unit time per bit or chip of the memory, and the SEU cross-sectional area σ _{SEU, b / c} (subscripts b and c represent bits and chips respectively) Correspondence) It is defined by the following equation.

[0003]

(Equation 1)

When the unit of SER is defined as bits or times / s per chip, the memory capacity of the chip is N _m (bits).
As

[0005]

(Equation 2)

As described above, the SEU sectional area is associated with _Sqc .

[0007] Low power consumption, high speed, and light weight due to the high integration of semiconductor memories are indispensable for creating added value, and the continuous flow of high integration and miniaturization is inevitable in the future. It is thought that there is. In the 1980s,
U and in the package material, soft errors due to α rays naturally occurring, such as ²³² Th was manifested as a constraint for the miniaturization. α-rays (bare nuclei without two orbital electrons of ⁴ He atoms) ionize Si atoms and the like in the memory element to generate electron-hole pairs (carriers).
It has been found that these carriers diffuse in the device, change the amount of electric charge stored in the electrodes in the memory, change the contents of the memory, and in some cases, lead to a latch-up phenomenon.

On the other hand, at present, soft errors due to α rays are almost completely solved by the following measures.

(A) Purification of package material (b) Increase / secure of memory capacity by adopting stack structure etc. (c) Error Correction (ECC)
Installation of ion codes) With the further integration and miniaturization of devices using semiconductor memory and the reduction of memory capacity, soft errors due to cosmic-ray neutrons, which have already become a problem for space rockets and satellites, are on the ground. Have also begun to be seen as problematic (eg, CAGossett et al., IEEE Trans. Nuclear S
cience, Vol. 40, No. 6, p. 1845 (1993), E. Norma
nd, IEEE Trans. Nuclear Science, Vol. 43, No.
6, p.2742 (1996), K. Johansson et al., IEEE Trans. N
uclear Science, Vol. 45, No. 3, p. 1628 (1998), T.
J. O'Gorman, IEEE Trans.Electron Devices, Vol.4
1, No.4, p.553 (1994)). By the way, the cause of neutron generation is explained as follows. In other words, it takes about one billion years from the center of the galaxy
Light ion particles, such as H, He, C, and O, are flying into the Earth's atmosphere. The ion beam is atmosphere of nitrogen, cause oxygen and nuclear reaction, high energy mesons ^{(π +, π 0, π} -,
μ) and neutrons, which in turn cause a nuclear reaction with the atmosphere in a chain, resulting in the formation of a neutron shower over a radius of several hundred meters from the surface of a single ion. As shown conceptually in Fig. 1, the orbit of an ion beam is bent by the magnetic field formed by the galaxy and the sun, and its intensity is affected by the solar activity with an 11-year cycle. The cosmic rays from the sun itself are low on average, but the intensity of cosmic rays reaching the ground is weak at the peak of solar activity (Solar Maximum) due to the effect of the solar magnetic field due to the intensity of solar activity. Becomes stronger when activity is inactive (Solar minimum). Since the orbit of ions is also bent by the earth's magnetic field, the flux toward the earth is maximized at the magnetic poles of the earth (the ion orbit tends to go to the surface of the earth along the lines of magnetic force because the lines of magnetic force are perpendicular to the surface of the earth), and the magnetic latitude (geomagnetic rigidity ) Depends. In addition, the neutron flux is altitude-dependent, with a maximum of 10-20 km above the surface of the earth, due to the neutron shielding capacity of the atmosphere itself. Neutron differential energy vector dφ _n / dE at the surface of the earth (units in the figure are for cosmic ray neutrons, particles / cm ² / MeV / s, WNR (high energy / high flux neutron irradiation equipment at Los Alamos National Laboratory) ) Is one pulse / cm ² / MeV / str / proton) is obtained by actual measurement, and has a high energy up to a region exceeding 1 GeV as shown in FIG. Here, a soft error of a semiconductor device on the ground due to cosmic ray neutrons will be considered.

Neutrons do not react with Si or the like which is a constituent material of a device at low energy. However, cosmic ray neutrons cause soft errors because neutrons with very high energy react with the nuclei of Si and split into multiple secondary ions (referred to as spallation reactions).
It is assumed that the secondary ions generate electron-hole carriers in the device. Therefore, the mechanism of soft error generation is the same as that of the conventional α-ray soft error.However, it is difficult to take countermeasures due to high-energy neutrons, which are difficult to shield. Soft errors occur at the same time (multi-bit error, 2 simultaneous doubles, 3 simultaneous triples, 4 quads
ruple bit error). Since multi-bit errors cannot be repaired by ECC unlike the past,
Higher ratios can create new problems.

[0011] Soft errors caused by impurities of the α-emitter in the package material are easier to evaluate for resistance to soft errors than neutrons in the following points.

(1) Since the generation position and the energy are clear, the spatial range of influence in the device is clear.

(2) Since the particles are only α particles, soft error resistance can be evaluated only by contacting a commercially available α-ray source with the device surface.

(3) Since it is a problem of only the package, the test results at a specific laboratory before shipment are not dependent on regions and are basically accepted worldwide.

Conversely, to rephrase the above for cosmic-ray neutrons: (1) Nuclear reactions occur at any location in the device,
Energy also has a broad spectrum.

(2) Since all kinds of nuclides that are lighter than the elements constituting the device are generated according to the energy of the incident neutrons, a complete simulation in an experiment requires a large-scale neutron source.

(3) The intensity of cosmic-ray neutrons is determined by the latitude, longitude, altitude on the earth, and the intensity of solar activity. In the background described above, conventional experimental evaluation methods for semiconductor device soft errors caused by cosmic-ray neutrons include a case where neutrons are used directly and a case where protons which are the same nucleons as neutrons are used. State. (1) Method using protons Based on the assumption that protons and neutrons are equivalent in spallation reactions, protons are widely used instead of neutrons, and the soft error part depends on the kinetic energy of the incident protons. It has been shown that the cross-sectional area (SEU cross-sectional area σ (E) for the kinetic energy of a particular incident particle) is different. (JF Ziegler, IBM J. Res. Develop., V
ol. 40, No. 1, p. 51 (1996), GASAi-Halasz, IEEE I
EDM Tech, Digest, p.344 (1983)). (2) Method of using neutrons directly Since the target is cosmic ray neutrons, the most accurate method is to use cosmic ray neutrons directly. However, since the flux is low, it is usually 1000 to obtain sufficient statistics. A test period of several months is required after arranging about a few chips (TJ O'Gorman, IBM J. Res. Develop., Vol. 4
0, No. 10, p. 41 (1996)). If it is carried out at high altitude, the flux can be obtained 10 to 20 times, but it is necessary to manage the temperature and secure the power supply.

Reactor neutrons have an energy of at most several MeV
However, the flux of neutrons exceeding 10 MeV is low and the experiment is difficult, but the efficiency is low, and it is practically impossible to use it including the experimental reactor.

It is possible to generate a neutron using an accelerator and perform a simulation experiment. By using the DT reaction, a neutron beam can be obtained with a relatively small device, but the energy is
It is not possible to evaluate the effects of high energy neutrons only at 14 MeV. In order to obtain high-energy neutrons, it is common to use a proton ring cyclotron, and the LANL (Los Alamos National Laboratory) WNR (Weapons Neutron Research)
At the facility, a proton accelerated to 800 MeV is irradiated on the target of the W block to obtain a neutron beam having a spectrum close to cosmic-ray neutrons as shown in Fig. 2. From this experiment, the total SEU cross section σ _SEU was obtained. (A. Eto et al., IEDM, p.367 (1998), Y. Tosaka et al., I.
EEE Transactions on Electron Device, Vol. 45, N
o. 7, p. 1453 (1998), CA, Gossett et al., IEEE Trans. Nu.
clear Science, Vol. 40, No. 6, p1845 (1993)). The soft error rate in the real environment is calculated from the product of the total SEU cross section and the total flux of cosmic ray neutrons. As a recent method, a proton beam is applied to a ⁷ Li
Increasing the number of incident neutron beams using near-single energies using the ⁷ Li (p, n) ⁷ Be reaction has increased, and even for neutrons, the energy dependence of the partial SEU cross section has increased. (K. Johansson et al.,
IEEE Trans.Nuclear Science, Vol. 45, No. 3, p16
28 (1998), P. Hazucha et al., IEEE Trans. Nuclear Scienc
e, Vol. 45, No. 6, p. 2921 (1998)).

On the other hand, the following report has been made on a conventional theoretical evaluation method of a semiconductor device soft error caused by cosmic ray neutrons.

The simulation of the phenomenon after the cosmic ray neutrons are incident on the Si atoms of the device is performed by dividing the reaction into a pre-equilibrium process and an evaporation process, and obtaining the energy spectrum of the generated particles by the Monte Carlo method. Evaluate soft errors by using a method such as this. As described below, the pre-equilibrium process is based on the assumption that neutrons incident on a Si nucleus repeatedly scatter between the nucleus and the nucleus in the nucleus, and that those with high energy are emitted out of the nucleus. By law (HHK
Tang et al., Phys. Rev. C, Vol. 42, No. 4, p. 1598 (199
0), K. Chen et al., PM O'Neill et al., IEEE Trans. Nucle
ar Science, Vol.45, No.6, p.2467 (1998), etc., when solved by a quantum molecular dynamics method as a many-body problem (Y. Tosaka et al., IEEE Trans. Nuclear Scie
nce, Vol.45, No.7, p.1453 (1998)). The evaporation process is calculated by the Weisskopf-Ewing evaporation model.

[0022]

In order to ship a semiconductor device as a product, it is necessary to accurately evaluate the quality of the semiconductor device. Since the evaluation criterion is a soft error rate on a FIT basis, the quality of the semiconductor device in an actual environment is high. It is necessary to obtain numerical values with high accuracy. At the design stage, a method for estimating the soft error rate of a device and an accurate test method for a prototype in a short period of time are required, but there is no established method. In the case of soft errors caused by cosmic-ray neutrons, multi-bit errors are errors in modes that have not existed in the past, and are important, but there is no method for evaluating them by theoretical analysis.

If the SEU cross section has no energy dependence, σ _SEU is determined by analysis or experiment for a specific incident energy, and it is calculated from (Equation 1) and (Equation 2) from the total flux incident on the device. good.

However, in practice, as has been clarified by experiments, because of energy dependence, strictly speaking,
The minimum value of the kinetic energy K to be evaluated for cosmic ray neutrons is K _min , the maximum value is K _max , and the partial SEU cross section corresponding to K is σ (K),

[0025]

(Equation 3)

Therefore, it is necessary to evaluate by superposition integration of the cosmic ray neutron spectrum and the partial SEU cross section.
On the other hand, it is practically impossible to measure or analyze the detailed numerical values necessary for the integration of (Equation 3) for an individual device, and in fact, the evaluation value of the specific device for the soft error rate in the actual use environment Has no reports.

According to the field test using actual cosmic rays, there is no such problem, and a numerical value in the unit of FIT can be obtained directly. However, it is an inefficient method that requires about six months for one evaluation. is there. Although it is extremely important as basic data, the test method is a technique that is not applicable to recent semiconductor devices that are being manufactured in a large variety of small-quantity products and are being increasingly integrated at an accelerating rate. On the other hand, accelerator-based methods can collect data in a short period of time, but proton-based methods give three times the SEU cross section for neutrons of the same energy (K. Johansson et al., IEEE Trans. Nuclear Scien
ce, Vol. 45, No. 3, p. 1628 (1998)), which is not a well-established method and may give a large error.
And neutron irradiation experiments using a single energy is a powerful technique, but every time the energy is changed,
It takes time-consuming work to measure the spectrum of neutrons, and there is a difficulty in surveying energy as a detailed parameter.

An apparatus that can obtain a soft error rate in a real environment of a semiconductor device in a short period of time in units of FIT is WNR, and as described above, its use is increasing in recent years. However, according to the inventors' theoretical analysis described below, WN
Neutron irradiation experiments using R revealed that the estimated soft error for cosmic-ray neutrons could be overestimated by 50%. Therefore, at least at present, there are no facilities in Japan and overseas that allow semiconductor device manufacturers to measure their own devices at the required speed and quantity.

An object of the present invention is to provide an accurate evaluation value of a soft error rate including a discrimination of a multi-bit error in a real environment of a semiconductor device caused by cosmic ray neutrons in a semiconductor device design / prototype stage. And to provide a process for reducing soft errors due to cosmic rays in a new device before mass production.

[0030]

According to the present invention, the spectral distribution of cosmic-ray neutrons is divided into energy bands having a predetermined energy value as a representative value, and the soft error partial cross-sectional area of the semiconductor device corresponding to the energy is divided. The partial cross-sectional area is weighted by the total flux for each energy band described above to obtain the sum of the soft error partial cross-sectional areas, and the soft error rate of the semiconductor device in the actual use environment is estimated based on the result. .

The soft error rate is obtained by setting the representative value of the above energy as the energy of incident neutrons in calculation, and calculating the secondary particles and residual nuclei generated by the nuclear reaction of the incident neutrons entering the inside of the semiconductor device. The amount and position of electron-hole pairs are calculated along the flight trajectory, and based on the calculation result, the amount is calculated from the amount of electron-hole pairs included in a predetermined region of the semiconductor device. I can do it.

Then, this flight trajectory is determined by using at least one of the generation position of nuclear reaction by incident neutrons, the scattering direction of secondary particles and residual nuclei generated by the nuclear reaction, and the kinetic energy, momentum, or velocity thereof. Let's calculate.

Further, according to the present invention, the cosmic ray neutron spectrum distribution is divided into energy bands having predetermined energy values as representative values, and corresponding to the total sum of the fluxes contained in each of the divided energy bands, The semiconductor device is irradiated with neutrons having the above representative values of energy, and a soft error rate corresponding to the cosmic rays of the semiconductor device at that time is experimentally obtained to evaluate its resistance.

By simultaneously irradiating neutron beams having different energies to the individual semiconductor devices, a partial cross-sectional area of a soft error generated in the semiconductor device is actually measured. To evaluate. Further, in the present invention, it is determined that a soft error has occurred in a semiconductor device when the amount of electron-hole pairs included in a predetermined region of the semiconductor device exceeds a predetermined threshold.

Further, the representative energy corresponding to the above energy band is set to an average value weighted by the energy differential flux in the energy band of the cosmic ray neutron spectrum, and the total flux for each energy band is set to be equal. I do.

According to the present invention, there is provided an ion beam generator, a target, a soft error measurement assembly in which a semiconductor device is arranged, and an analyzer for calculating a soft error rate of the semiconductor device. Then, this analyzer divides the spectral distribution of cosmic ray neutrons, sets a representative energy value in the divided energy band as neutron energy for irradiation to the semiconductor device, and irradiates the semiconductor device with neutrons. After that, the control means is controlled so as to read out the recorded content of the semiconductor device, compare it with the previously recorded content, and use the result to determine whether or not a soft error has occurred in the semiconductor device. I made it.

The chamber containing the target has at least two outlets through which a neutron beam generated by a nuclear reaction between the target and the ion beam emitted from the ion beam generator is emitted. The measurement assembly was arranged on the axis of the neutron beam line at each outlet, and the neutron beam was irradiated simultaneously.

[0038]

Embodiments of the present invention will be described below in detail with reference to the drawings.

In this embodiment, the execution of the superposition integral of (Expression 3) is impossible for both analysis and experiment. Therefore, after the spectrum of the cosmic ray neutrons is discretized, We propose a method to evaluate the soft error rate. The basic concept is the same for both theoretical analysis and experiment, and a basic evaluation analysis flow is shown in FIG. Hereinafter, the evaluation procedure will be described.

(1) First, the cosmic ray neutron spectrum distribution is divided into several energy bands, and a representative energy K _i ^* for each band is determined by, for example, the following equation.

[0041]

(Equation 4)

If the band width is sufficiently small, a simple average may be used.

Although the upper and lower limits of the corresponding energy band may be arbitrarily determined, here, the total flux of cosmic ray neutrons

[0044]

(Equation 5)

On the other hand, according to the following equation, the flux in each band is equal to each other.

[0046]

(Equation 6)

From K _min (i = 1) in order so that
It will determine the K _i.

(2) Irradiate a neutron beam having the above representative energy K _i ^* to a target semiconductor device.

(3) Thus, the representative energy K
_{When i} ^* neutrons are incident on a total of N _n devices, the average is n
Assuming that (K _i ^* ) soft errors have occurred, the SEU partial cross-sectional area σ _i (K _i ^* ) of the device is obtained by the following equation.

[0050]

(Equation 7)

(4) Using this, the predicted soft error rate SER for cosmic ray neutrons of this device can be calculated as follows.

[0052]

(Equation 8)

The total SEU cross-sectional area σ _cosmic is as follows, and this parameter makes it possible to compare the soft error resistance of various devices due to cosmic ray neutrons.

[0054]

(Equation 9)

Next, a method for predicting a soft error rate caused by cosmic ray neutrons based on a conceptual design of a semiconductor device will be described.

FIG. 4 shows the overall flow of the analysis.

First, as schematically illustrated in FIG. 5, the memory portion of the semiconductor device has a width of 2W _x , a depth of 2W _y , and a height of T.
Set bulk Si block of _w . The size of the block corresponds to the interval between bits of the memory cell, the thickness of the film formation layer, and the like. Inside the block, the radius r _s corresponding to the memory bits,
Assume a spherical sensitive area with a height d _s at the center from the bottom. The sensitive area corresponds to the bit or capacitor, r _s ,
Is related to its size, and _ds is related to the thickness of wells, depletion layers, substrates, etc. Then, the particles pass through a region provided in the semiconductor device, and the charge generated inside is Q _c
If a value exceeds the limit, a soft error will occur.

The number of the above regions is set to four so that multi-bit errors can be evaluated. The position at which the incident neutrons cause the spallation reaction is determined randomly within the block.

Next, cosmic ray neutrons and protons
In order to evaluate neutron irradiation experiments widely, Monte Carlo calculations are performed with a weighted spectrum, assuming a wide energy spectrum of incident nucleons. Specifically, FIG.
As shown in the example, the number of incident neutrons to be calculated is N _Total , the minimum energy is K _min , and the maximum energy is K _max
, Set N _b fine energy bands between K _min and K _max , and set the energy of the ith band to the minimum K
_i-1, the maximum K _i. The representative energy K _i ^* in the band is represented by the following equation, for example, as a simple average.

[0060]

(Equation 10)

To express more accurately, using an average weighted by the differential flux of the cosmic neutron beam,

[0062]

[Equation 11]

## EQU5 ##

The total flux represented by the following equation

[0065]

(Equation 12)

On the other hand,

[0067]

(Equation 13)

[0068] As will be, will determine the sequence K _i from K _min (i = 1).

Here, for each of the bands,
Consider the case of the average energy -K _i ^* of the neutron incident N _{Tot al} / N _b times.

The upper and lower limits of the bands may be arbitrarily determined. In this case, the number of neutron incidences for each band is determined in proportion to the sum of the flux for each band.

Next, a description will be given of the import of numerical values from the database.

After the energy of the incident particles and the target material are determined, data necessary for the soft error simulation is fetched from the database. The types of data include an atomic number, a mass number, a static energy, a spin, and a nuclear reaction cross section of the target material and all nuclei and nuclei that can be generated from the target material.

Next, from the above-mentioned data, analysis from neutron incidence to nucleon emission and generation of excited residual nuclei will be analyzed.

The nucleus is first released from the Si nucleus in the device into which the neutron is incident by a pre-equilibrium process. Pre-equilibrium process of spallation reaction (10 ^-22 to 10 ^-18 s after injection before complex nucleation)
Phenomenon) is a process in which the incident nucleon sequentially excites the nuclear nucleus and a part of the nucleus having high energy is emitted outside the nucleus, as shown in FIG. There is
Calculations consider this as a cascade of two-body elastic scattering of nucleons in the target nucleus (Intra-Nuclear Cascade), and analyze the motion of the nucleons in the target nucleus in order by Monte Carlo method, Calculate the kinetic energy, direction, and excitation energy, kinetic energy, and direction of the remaining nuclei (residual nuclei) after the nucleon is released. The excitation energy of the remaining nuclei is calculated as the sum of the kinetic energies of the nucleons that are not emitted out of the target nucleus among the nucleons scattered by the Intra-Nuclear Cascade.

The residual nucleus having this excitation energy is Wei
According to the evaporation theory of sskopf-Ewing, light particles (n, H, H
e) release. The type of particles emitted, energy,
The direction is calculated by the Monte Carlo method based on the reaction cross section data.

Next, the trace of the emitted particles when flying in the semiconductor device will be analyzed.

As a result of the spallation reaction, all nuclei lighter than Si, including Si, can be generated. Each of these nuclei scatters in the virtual Si substrate with its own spectrum and angular distribution, and loses energy along the track mainly due to Coulomb interaction with the electron cloud. This energy loss dK / dx is a function of the kinetic energy of the particle,
It can be calculated by Bloch's formula. Therefore, if the value of dK / dx is calculated along the track, the initial energy _Kq0
Fei as R _q of the _particles, as a result, the stop position of each particle can be calculated.

[0078]

[Equation 14]

Here, the amount of generated carriers is calculated in a region provided in advance in the semiconductor device.

This carrier is formed by using the energy q _p (3.6 eV in the case of Si) necessary for forming an electron-hole pair.
(DK / dx) / q _p is the electron-hole pair density generated along the track. When the particle stops in the region, the calculation is performed on the assumption that all of the initial kinetic energy of the particle in the region is used for carrier generation. Seeking carrier generation density along the track of the particle Otherwise, calculate the total charge amount generated in sensitive areas, it determines it if more than Q _c, a soft error occurs. If a soft error occurs in a plurality of areas for one neutron injection, it is considered that a multi-bit error has occurred.

In FIG. 8, when 1000 neutrons with an energy of 105 MeV are incident, W _x = W _y = d _s = 2 μm and T _w = 5 μ
m, for the device of Q _c = 10 fC, illustrates a calculation example of the soft error count when changing the radius of the area. It is shown that the analysis model enables detailed evaluation of the multi-bit error rate after changing various device parameters.

The calculation of the soft error rate generated in the semiconductor device can be obtained as follows.

Assuming that the device installed on the neutron beam line corresponding to the representative energy K _i ^* and the flux has caused an average of n (K _i ^* ) soft errors per chip during the time T, the SEU of the device Partial cross section σ _i (K _i ^* )
Can be calculated by the following equation.

[0084]

(Equation 15)

Using this, the predicted soft error rate SER for cosmic ray neutrons of this device is as follows.

[0086]

(Equation 16)

The total SEU cross-sectional area σ _cosmic is obtained by the following equation, and this parameter makes it possible to compare the soft error resistance of various devices due to cosmic ray neutrons.

[0088]

[Equation 17]

The irradiation time T _irrad in ( _Equation 15) is the irradiation time in the actual environment, and is calculated from the volume V of the device.
Number of particles N incident on _devices (cm ³ ), scattered and reacted
Calculated from _total (number). Atomic number density N _{tar of} target material
(Pcs / cm ³ ) means ρ _tar and M _tar respectively the density of the target material
(g / cm ^3), the mass number, the N _A as Avogadro number, as follows.

[0090]

(Equation 18)

The differential flux of incident particles that is actually assumed is represented by the following equation.

[0092]

[Equation 19]

From this, the actual irradiation time T _irrad (s) is obtained by the following equation.

[0094]

(Equation 20)

Next, an experimental evaluation method of cosmic ray neutron soft errors for semiconductor devices will be described. Using the theoretical model described above, 2μm spacing in the semiconductor device,
Consider a case where a region having a radius of 0.5 μm and a height of 2 μm at the center from the bottom is provided.

The neutron spectrum of K _n = 1 to 1000 MeV is divided into 100 energy bands, and the calculated value of the soft error rate when 100 neutrons are incident on each band is defined as the cosmic ray neutron spectrum and WNR. Was performed on the spectrum.

As shown in FIG. 9, the neutron spectrum of the WNR was found to be overestimated by about 50% with respect to neutrons of cosmic rays.

The spectra of cosmic-ray neutrons in WNR and in nature are certainly similar, but our calculations clearly show for the first time that the experimental results of the soft errors of both are significantly different. . The difference is that the average energy of neutrons above 1 MeV is 46 MeV for cosmic rays and 77 MeV for WNR,
The direct reason is that the NR is nearly twice as high. The inventors' analysis has revealed for the first time that there is a problem with the evaluation method using irradiation having a broad spectrum such as WNR. Therefore, as a method of experimental evaluation, a test method using a single-energy neutron beam was devised based on the basic concept. That is, the neutron energy K _i ^* and the upper and lower limits of the spectrum band are calculated by the above (Equation 4).
Was set so as to satisfy the relationship, and a neutron irradiation test was performed. In the case of an irradiation test, the number of neutrons incident on each band may be arbitrarily determined. In this way, the device installed on the neutron beam line 5 corresponding to the representative energy K _i and the flux Φ (K _i ) causes an average of n (K _i ) soft errors per chip per time T. Then, the SEU cross section σ _i (K _i ) of the device can be expressed by the following equation.

[0099]

(Equation 21)

In the above formula, T _irrad differs from the analysis, and the irradiation time in the experiment may be used. From this, the predicted soft error rate SER and the total SEU cross section σ _cosmic of this device for cosmic ray neutrons are as follows.

[0101]

(Equation 22)

[0102]

(Equation 23)

By doing so, it becomes possible to compare the soft error resistance caused by cosmic ray neutrons among various devices using this parameter. As described above, the basic concept of the method for evaluating the soft error rate caused by cosmic-ray neutrons in the actual use environment of an arbitrary device has been described through analysis and experiments. It is necessary to ensure that the resistance of a semiconductor device to cosmic ray neutrons meets a standard before mass production of the device. In this embodiment, an example in which the quality of such a device is improved and secured will be described with reference to FIG. . From the parameters _Di (i = 1,2,) related to the characteristics and structure of the device in the conceptual design of a new device, the model parameters F _j ( j = 1,2,), and the soft error rate in the real environment is calculated by simulation based on this.

[0104]

(Equation 24)

When the calculated value of the soft error rate does not satisfy the standard, the model parameters are finely adjusted so as to satisfy the standard.

From the model parameters, device parameters are obtained by using the inverse correlation equation expressed by the following equation (Formula 24) and by performing a detailed device simulation as necessary to confirm the operation of the device. The prototype of the device is fabricated, and the soft error rate in the real environment is obtained by irradiation experiments with high energy protons or neutrons.

[0107]

(Equation 25)

[0108] For similar devices, once the evaluation value of the real environment is obtained, it is used as a standard device.
Performing the relative evaluation using only specific energy and proceeding with the optimization has the effect of reducing the time.

If the irradiation test result is significantly different from the prediction, the model parameters are readjusted and the process returns to the evaluation by analysis.

As described above, by using the evaluation method shown in the present embodiment, it is possible to predict the occurrence of soft errors due to cosmic ray neutrons of a semiconductor device before starting mass production of the semiconductor device. Thus, the risk of device development can be reduced.

FIG. 11 conceptually shows the structure of an evaluation device for explaining another embodiment.

The apparatus includes a soft error measuring assembly 18 in which a semiconductor device is arranged, and an analyzer 20 for calculating a soft error rate of the semiconductor device. The measuring assembly 18 is arranged on the axis of the neutron beam line 14. I have. Further, the analysis device 20 divides the spectral distribution of cosmic ray neutrons, and sets a representative energy value in the divided energy band as irradiation neutron energy to the semiconductor device disposed on the axis of the neutron beam line 14. Means for setting, and after the irradiation of the neutrons, read the recorded content of the semiconductor device,
It comprises a means for comparing with the pre-recorded contents, and a control means for performing control so as to determine that a soft error has occurred in the semiconductor device when the comparison result exceeds a predetermined value.

The protons and α particles accelerated by the ion beam generator 10 are extracted through the ion beam line 11. Then, this ion beam is applied to the target chamber 12.
The neutrons are generated by being incident on the target 13 disposed inside. At this time, when the angle of the neutron beam line 14 is the same as the direction of incidence of the ion beam, the energy of the generated neutron shows the maximum value, and the larger the angle made with the ion beam 11, the smaller the energy of the generated neutron.

In FIG. 11, the representative energy K _i ^* (i = 1,5) is shown on the axis of the neutron beam line 14, but K ₁ ^* ,
The energy increases in the order of K ₂ ^* . If the angle is constant, the reduction rate of the representative energy is constant, and the flux ratio is also constant, so that the neutron beam line 14 is provided radially around the target, for example, only the beam line in front of the ion beam. If the neutron spectrum is measured in the above, the other beam lines need only be measured once, and thereafter, the measured value of the front line may be multiplied by a coefficient. The energy spectrum is measured by a simple measuring device 15, such as a Bonner ball assembly, and a time-of-flight measuring line 1.
The measurement is performed using a highly accurate method such as No. 6.

The irradiation room 17 may be a continuous room. However, in order to eliminate the influence of scattered neutrons, the measurement assemblies 18 need to provide sufficient shielding 31.
In addition, since the operation of the device changes under the influence of temperature and humidity, the atmosphere in the irradiation chamber 17 needs to be controlled to a predetermined value by the temperature and humidity adjusting device 21. In addition, even if the ion beam is stopped, the generated neutrons are scattered in the irradiation chamber 17 with a half-life of about 10 minutes.
It is preferable that the radiation intensity in the irradiation chamber 17 is monitored by 2 or the like, and there is an interlock such that the door 23 in the irradiation chamber 17 is locked until the radiation intensity drops below a certain level. Since fluctuations in the power supply voltage cause a large error, it is desirable that the large-capacity stabilized power supply 24 be used together with a backup power supply using the emergency power supply 25.

FIG. 13 shows another embodiment of the present invention with respect to the target chamber 12. Of the neutron beam lines 14 extending radially from the target chamber 12, a beam dump 26 is provided in front of the ion beam line 11. The neutron beam line 27 having the maximum energy is arranged so as to be slightly shifted from the front. When the neutron beam line with the maximum energy is set in front of the ion beam line, the beam dump 26 is required in any case.
Since neutrons generated by the beam dump 26 themselves become noise, it is desirable to avoid them.

FIG. 12 shows a chamber 1 in which a target 13 is placed.
FIG. 2 illustrates a schematic diagram of FIG. In shamba 12,
At least two outlets are provided for taking out neutrons generated by a nuclear reaction between the ion beam and the target 13, and a semiconductor to be evaluated is provided on an extension of this outlet and on the axis of the neutron beam. A measurement assembly 18 containing the device is arranged. In this figure, the target 13 is tilted with respect to the incident ion beam, in order to avoid unnecessary scattering of neutrons emitted in a direction perpendicular to the incident ion beam.

As described above, during the time T, the number of semiconductor devices installed on the neutron beam line 14 corresponding to the representative energy K _i and the flux Φ (K _i ) is n (K _i ) on average per chip , The SEU cross section σ _i (K _i ) of the device is expressed by the following equation.

[0119]

(Equation 26)

Using the result, the predicted soft error rate SER and the total SEU cross section σ _cosmic of the device for cosmic ray neutrons can be obtained as follows.

[0121]

[Equation 27]

[0122]

[Equation 28]

As a result, the evaluation of the soft error resistance due to cosmic ray neutrons in various devices can be performed dramatically more efficiently than in the conventional method.

[0124]

According to the present invention, regarding the soft error resistance of a semiconductor device due to cosmic ray neutrons, it is possible to predict the soft error rate corresponding to the real environment in a short time.

[Brief description of the drawings]

FIG. 1 is a diagram for explaining the cause of cosmic-ray neutron generation and its influencing factors in an example.

FIG. 2 is an explanatory diagram showing differential fluxes of cosmic ray neutrons in Tokyo and neutrons in WNR.

FIG. 3 is a diagram for explaining an evaluation analysis flow of a soft error rate caused by cosmic ray neutrons in the present embodiment.

FIG. 4 is a diagram for explaining an analysis flow for calculating a soft error rate by calculation in the embodiment.

FIG. 5 is a schematic diagram for explaining a device model used for theoretical analysis in the present embodiment.

FIG. 6 is a diagram for explaining a method of dividing a differential flux of cosmic ray neutrons in the present embodiment.

FIG. 7 is a diagram for conceptually explaining an intranuclear scattering cascade of a spallation reaction in the present embodiment.

FIG. 8 is an explanatory diagram showing a relationship between a theoretical analysis result on the number of soft errors in the present embodiment and a region provided in a semiconductor device.

FIG. 9 is a diagram showing a comparison of the frequency of occurrence of soft errors when the same number of neutrons are incident on neutron spectra of cosmic ray neutrons and WNR in the present embodiment.

FIG. 10 is a diagram for describing an application example of the present embodiment.

FIG. 11 is a schematic diagram showing another embodiment.

FIG. 12 is a diagram illustrating an example of arrangement of irradiation neutron beams in another embodiment.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 10 ... Ion beam generator, 11 ... Ion beam line, 12 ... Target chamber, 13 ... Target, 14 ... Neutron beam line, 15 ... Simple neutron spectrum evaluation device, 1
6 neutron spectrum precision evaluation device, 17 neutron irradiation room, 18 soft error measurement assembly, 19 analysis room, 20 analysis device, 21 temperature / humidity adjustment device, 22
... neutron level monitor, 23 ... irradiation room door, 24 ... stabilized power supply, 25 ... emergency power supply, 26 ... beam damper, 27
… Neutron beam line

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Akira Eto 5--20-1, Josuihoncho, Kodaira-shi, Tokyo Within the Semiconductor Group of Hitachi, Ltd. (72) Inventor Hiroshi Kikuchi Yoshida-cho, Totsuka-ku, Yokohama-shi, Kanagawa Prefecture 292 F-term in Hitachi, Ltd. Production Research Laboratory (reference) 2G003 AA07 AA08 AG06 AH01 AH02 AH05 2G032 AA07 AB17 AC03 AC08 AE08 AE12 AE14 AG10 AH04 AK01 9A001 BB05 KK15 KK31 KK54 LL08

Claims (15)

[Claims] A step of dividing a spectral distribution of cosmic ray neutrons into an energy band having a predetermined energy value as a representative value; a step of obtaining a soft error partial cross-sectional area of the semiconductor device corresponding to the energy; Weighting the partial cross-sectional area with the total flux for each energy band to obtain a sum of the soft error partial cross-sectional areas, and using the sum of the soft error partial cross-sectional areas to obtain the semiconductor device in an actual use environment. A method for evaluating cosmic ray soft error resistance of a semiconductor device, comprising estimating a soft error rate of a semiconductor device. 2. The secondary particle and residual nucleus generated by a nuclear reaction of the incident neutron entering the inside of the semiconductor device, wherein the soft error rate sets a representative value of the energy as the energy of the incident neutron in calculation. The amount and position of electron-hole pairs are calculated along the flight trajectory, and based on the calculation result, the amount of electron-hole pairs included in a predetermined region of the semiconductor device is calculated from the amount of electron-hole pairs. 2. The method for evaluating cosmic ray soft error resistance of a semiconductor device according to claim 1, wherein the error rate is calculated. 3. The flight trajectory of the secondary particles and the residual nuclei, the position of the nuclear reaction caused by the incident neutrons, the scattering direction of the secondary particles and the residual nuclei generated by the nuclear reaction, and the secondary particles and the residual nuclei. 3. The method for evaluating cosmic ray soft error resistance of a semiconductor device according to claim 2, wherein the calculation is performed using at least one of kinetic energy, momentum, and velocity of the semiconductor device. 4. A cosmic ray neutron spectrum distribution is divided into energy bands each having a predetermined energy value as a representative value, and the energy of the energy is divided according to the total flux contained in each of the divided energy bands. The method for evaluating cosmic ray soft error resistance of a semiconductor device according to claim 1, wherein the semiconductor device is irradiated with a neutron having a representative value. 5. The semiconductor device according to claim 1, wherein when a generation amount of said electron-hole pairs included in a region of said semiconductor device exceeds a predetermined threshold value, it is determined that a soft error has occurred in said semiconductor device. A method for evaluating cosmic ray soft error resistance of a semiconductor device according to claim 2. 6. A method according to claim 1, wherein a plurality of regions are provided in advance in said semiconductor device, and a generation amount of said electron-hole pairs generated in at least one of said regions exceeds a predetermined threshold value. 3. The method according to claim 2, wherein it is determined that a soft error has occurred in the semiconductor device. 7. The soft error partial cross-sectional area generated in the semiconductor device by irradiating neutron beams having different energies to the individual semiconductor devices at the same time. Cosmic ray soft error tolerance evaluation method for semiconductor devices. 8. The semiconductor device according to claim 7, wherein the neutron beam is formed by absorbing a high energy ion emitted from the accelerator and using a thin plate for generating neutrons as a target. Cosmic ray soft error tolerance evaluation method. 9. The method for evaluating cosmic ray soft error resistance of a semiconductor device according to claim 8, wherein when said high energy ions are protons, ⁷ Li is used as a target. 10. When the high energy ion is ⁴ He, The method for evaluating cosmic ray soft error resistance of a semiconductor device according to claim 8, wherein the method is used as a target. 11. The universe of a semiconductor device according to claim 1, wherein the representative energy corresponding to the energy band is an average value in the energy band of the cosmic ray neutron spectrum weighted by an energy differential flux. Line soft error tolerance evaluation method. 12. The method for evaluating cosmic ray soft error resistance of a semiconductor device according to claim 1, wherein the total flux for each energy band is set to be equal. 13. An ion beam generator, a target, a soft error measurement assembly having a semiconductor device disposed thereon, and an analyzer for calculating a soft error rate of the semiconductor device, the analyzer comprising a cosmic ray neutron. Means for setting the representative energy value in the divided energy band as the neutron energy for irradiation to the semiconductor device, and means for recording and reading the recorded contents of the semiconductor device. Means, having control means, after the irradiation of the neutrons is performed by the comparing means, read out the recorded content of the semiconductor device, compare with the pre-recorded content, the comparison result is predetermined The control means is controlled to determine that a soft error has occurred in the semiconductor device when the value exceeds the value. Soft error tolerance evaluation apparatus of a semiconductor device to. 14. A chamber storing said target, comprising at least two outlets through which a neutron beam generated by a nuclear reaction between said target and an ion beam emitted from said ion beam generator is emitted. 14. The apparatus for evaluating soft error resistance of a semiconductor device according to claim 13, wherein: 15. The soft error measurement assemblies are respectively arranged on axes of neutron beam lines having different energies, and the control means is controlled such that the neutron beams are irradiated simultaneously. The apparatus for evaluating soft error resistance of a semiconductor device according to claim 13.