3.2.1 Components of Bragg Cell Receivers: Light Sources

The light source is a key component in the Bragg cell receiver.

The Bragg receiver requires a collimated, monochromatic (coherent), low-noise source for optimum performance. In the optical processor, the light source is like a local oscillator in superheterodyne receiver. Noise on the oscillator will be transferred to the output. Helium-Neon (HeNe) lasers are ideally suited for this application, possess narrow spectral bandwidths, and have demonstrated lifetimes over 10,000 hours. LED’s can be used as sources in incoherent optical processors, but are not suitable for the simple Bragg cell receiver, because of their broad spectral emission. Semiconductor lasers have spectral bandwidths narrower than most LED’s but not as narrow as a HeNe laser. For applications where high resolution is not required, semiconductor lasers may even improve performance because of their high output power with a much smaller size and lower power consumption. [10]

For some materials such as LiNbO₃, the AO diffraction efficiency using an 850 nm source can be reduced to 50% of the efficiency using a 633 nm source [11]. The higher output power of the semiconductor laser offsets this. While the choice of the source depends upon the application, for the ultimate in performance, where size and power consumption are less important, the HeNe laser is used. Where miniaturization and power consumption are most important, the semiconductor laser is the best choice. The semiconductor laser must be butt coupled to the edge of the IO Bragg cell. Materials that make a good laser, such as AlGaAs, do not possess good AO properties, and the good AO materials, such as LiNbO₃, cannot be made into lasers.

3.2.2 Lenses

The lenses in an IO Bragg cell are waveguide lenses. Various types of waveguide lenses have been devised, four of which are illustrated in Figure 3.6 [12]. The same effect as conventional lenses is created with step increases in thickness. A dome-shaped thickness increase is a Luneburg lens; and the inverse, a dome-shaped depression, is basically a geodesic lens. The geodesic lens is the most highly developed, because it is easier to grind the waveguide to near spherical depression than to add material to create a dome. Holographic lenses can be produced by creating grating structures with varying periodicity. The geodesic lens functions using the geometrically longer paths of rays in the center portion relative to the edges. Two lenses are used in the Bragg cell receiver—one between the light source and the cell and the other between the cell and the photodiode array.

The beam collimator determines the interaction aperture of the acoustic wavefront. The wider the aperture, the more frequency resolution is possible, to a point. The acoustic attenuation of various materials limits the aperture size.

Several mechanisms are thought to be responsible for acoustic attenuation. In many materials, Akheiser loss caused by thermal relaxation is the dominant mechanism. For these materials, the attenuation (in dB per unit length) will increase as the square of the frequency. Other acoustic modes exhibit frequency dependencies that range from linear to square. For these materials, other mechanisms such as scattering from lattice imperfections are important in the attenuation process. The acoustic attenuation measured for certain acoustic modes tends to vary widely from sample to sample, even for pieces cut from the same crystal boule. This indicates that lattice dislocations and foreign particle impurities are dominant in the attenuation process. While measures of these crystal imperfections such as etch pit density have been correlated to optical and electrical crystal properties, these measures have not been applied in acoustic attenuation studies. [13]

Figure 3.7 compares several acoustic modes which exhibit low acoustic attenuation [14].

3.3.1 Fourier Transform, Fourier Transform Lens

The function of the Fourier Transform lens is to rearrange the diffracted light beam output from frequency-dependent Bragg angles to frequency-dependent positions, so that the output of the PD array is a linear distribution of frequencies. The light output, or what the PDs see is three sinc functions [18]:

u(xf)=k[sinclxfλF+m2sincl((xfλF-f)+m2sincl(xfλF+f))]⁢(3.4)

where

x_f is the position of the spot

l is the AO aperture

m is the grating modulation depth or RF signal level

f is the reciprocal of the diffraction grating spacing or 1/Λ

k is a constant that accounts for the laser light output and the system’s optical losses

F is the distance between the lens and the focal plane or PD array

The first term is the zeroth order term which is not dependent on the RF input, and the last two terms are the first order terms, the “−f” term being the downward Doppler shift, and the “+f” term the upward shift. Figure 3.10 is an illustration of this.

3.3.2 Dynamic Range

The light intensity on the PD array is proportional to the square of the intensity:

l(xf)=|U(xf)|2⁢(3.5)

This illustrates Bragg cell technology’s weakness: dynamic range. For example, for a Bragg cell receiver to have 60 dB of RF dynamic range, the PD array must have 120 dB. Dynamic range will be referred to with the RF signal as the reference. One hundred and twenty decibel (60 dB RF) is beyond the performance for most PDs. The PD limits the RF dynamic range of the system to between 25 and 40 dB, depending on the type of PD used, but the Bragg cell itself has a dynamic range between 50 and 70 dB [19].

The two categories of linear array PDs useful for AO signal processing are the photodiode and the MOS depletion mode sensor [20].

One example of a commercially available randomly addressable photodiode linear array is the Reticon CP1006 device. It consists of two interdigitated rows of 256 photodiodes with a width of 50 μm. Its dynamic range is 20 dB for noncoherent (conventional) detection processors. Another commercially available array is the Reticon CP1023 device, with CCD shift registers. It has 256 elements on an 18 μm pitch, with four independent outputs of 64 elements each, capable of a 6 MHz data rate (10.6 μs access time). Its specified dynamic range is 30 dB for noncoherent detection systems. A detector that has been built for AO Braggs cell receivers has 140 elements on a 12 μm pitch, including an element for zeroth order dumping. The measured dynamic range is 41 dB for 2 μs access time. A MOS-CCD PD has been built to be used with an IO Bragg cell receiver with four groups of 25 elements each. It is designed to be read out in four parallel channels, or two parallel channels of 50 elements.

No dynamic range data was given. Both types of PDs are shown in Figure 3.11 [21].

3.10.1 Introduction

All optical signal processing systems for both optical communications and optical computing have tremendous potential for dramatically improving the information handling capacity and data rate in various signal processing systems by keeping the information in optical form during most of the processing path [33]. Major improvements in real-time signal analysis can be obtained through the implementation of all-optical circuits due to the advantage of low power consumption, high speed, and noise immunity. Additionally, all optical systems lend themselves directly to implementation in highly parallel, high throughput architectures.

Dynamic nonlinear optical processes lie at the heart of these optical logic devices. The index of refraction of a semiconductor can be significantly varied through the creation of electrons, holes, and excitons or by exposure to high intensities of light. Changes in these properties can induce absorptive or dispersive bistability in the semiconductor system [34]. A system is said to be optically bistable if it has two output states for the same value of the input over some range of input values. Relaxation properties of the excited semiconductor state dictate the characteristics of these optically induced optical changes.

Gibbs et al. [35] in 1979 were among the first to apply excitonic effects to optical bistability in GaAs. Bistability was demonstrated just below the exciton resonance in a cryogenically cooled, 4 μm thick, GaAs sample utilizing approximately 200 mW of optical power. The device exhibited a switch-off time of approximately 40 ns. This bistability was due to the change in refractive index resulting from nonlinear absorption in the vicinity of the exciton resonance. One limitation to further progress in this area was that excitonic resonances are generally not seen in room-temperature direct-gap semiconductors due to temperature broadening effects.

It has been found, however, that multiple quantum well (MQW) structures show strong exciton resonances at room temperature due to the QW confinement and consequential increase in binding energy of the excitons [36]. These resonances show saturation behavior similar to that observed at low temperature in conventional semiconductors. As such, the MQW structure represents a preeminent class of device to be researched for implementation in optical logic function units.

Bulk and integrated bistable devices have demonstrated the ability to perform a number of logic functions [37]; however, various device parameters such as the physical size, power requirements, and speed and temperature of operation have generally proved to be impractical. Two devices that show promise as optical processing components are the Double-Y Mach–Zehnder Interferometric Optical Logic Gate (Double-Y) and the MQW Oscillator.

3.10.2 Interferometer and Quantum Well Devices

Individual MZIs have been demonstrated in various materials and have been used as modulators and switches [38]. The MQW Oscillator makes use of the nonlinear optical effects unique to the MQW structure. Both the Double-Y, which is an extension of the MZI concept, and the MQWO devices lend themselves directly to fabrication in GaAs material. Both devices can overcome the limitations of speed and operational temperature mentioned previously. In its present configuration, the Double-Y is about 2 cm long and requires rather large optical powers for switching. On the other hand, the MQW is already a small, low-power device.

The Double-Y and MQW oscillator devices utilize a total of three optical effects to perform their logic and switching functions. These are the linear electro-optic, or Pockels effect, the nonlinear index of refraction, and the Quantum Confined Stark Effect (QCSE). Derivations of the first two effects are included at the end of Chapter 2. The Double-Y device has been studied and fabricated in LiNbO₃ [39] and in AlGaAs [40]. The MQW Oscillator can be implemented in several forms, one of which has been studied by Wood et al. [41].

A schematic diagram of the Double-Y device is shown in Figure 3.16. The signal of interest is input at the center or control leg. The signal splits at the Y and then recombines. The output depends on how this signal recombines. If the path lengths down the two legs are identical (or differ by an integral number of wavelengths) then the control signal recombines in phase and is referred to as a “one.” If the path lengths differ by an integral number of half wavelengths, the signal recombines destructively, resulting in a “zero” output.

The Double-Y device utilizes two nonlinear optical effects to change the path length of either or both legs and thus change the output state between “one” and “zero.” The Pockels effect provides a phase shift proportional to an electric field applied across the waveguide. This field is supplied by applying a voltage to the electrodes over the waveguide. The circuit is completed with a ground plane below the guide. These contacts are indicated in Figure 3.16. In the Double-Y device, this effect is used to tune the recombination of the signal. Tuning is needed to correct for any differences in the physical lengths of the two legs and to set the initial output state of the device to either “one” or “zero.” Calculations indicate that for the geometry shown in Figure 3.16, a phase change of 180° can be obtained by applying a potential of approximately 1.8 V.

The other effect used to change the path length is the nonlinear or intensity dependent index of refraction. By coupling a high intensity of light into either of the signal legs, “A” or “B,” the path length of the control pulse will be changed. It is seen, therefore, that after setting the initial state by using the Pockels effect, the relationship between the inputs “A” and “B” and the output will be exclusive—or (XOR) or inverse XOR logic functions, as described by the truth tables in Figure 3.16. One should note that the output polarizer selects only the parallel polarized control signal for the output. This has implications for the concept of cascading the basic Double-Y gate to implement other logic functions.

Several combinations of cascaded Double-Y type gates can be considered in order to implement all basic logic functions. Review of Figure 3.16a shows that three logic functions are accounted for by the basic Double-Y gate. Two of these are the XOR¯ and XOR functions. In addition, it can be seen that by using the gate in either of the above initial phase modes (i.e., in-phase or 180° out-of-phase) an inversion of the “A” input can be obtained by leaving the “B” input “off” or “on,” respectively. This is indicated by the dashed boxes in the truth tables.

Figure 3.16b shows a cascaded gate that produces either the NOR or AND function, depending on the setting of the B_⊥ input. A few of the possible problems that could be encountered in the operation of this gate are the splitting of the A input power and the length of the A_⊥ guide. Both of these factors can make the proper phase changes and synchronization difficult to obtain. These problems can be overcome with proper design control. Another problem is the extended length of the device; however, this would be reduced in accordance with reductions in the length of the single Double-Y gate.

Figure 3.16c shows the implementation of the OR and NAND functions. Although structurally more complex than the previous gates, it is a very symmetric design; and therefore, synchronization should be easier to obtain. On the other hand, the effects of the various splitters, combiners, and crossovers will need to be optimized.

A major problem with cascading the Double-Y logic gate is that an “on” output from a following gate cannot be generated by inversion of an “off” output from a previous gate. This is because the individual gates inherently use “on” control pulses. As these are progressively canceled, subsequent gates have fewer operations that can be performed. This is markedly different from electrical logic gates where a source to invert an “off” value is always available.

The MQW Oscillator of interest for optical computing is a hybrid device, which relies on an electrical signal to change the absorptive properties of the material. This change in absorption, caused by the QCSE [42] is a very important effect in MQW structures. These structures are crystals grown by alternating very thin layers of two materials which have similar lattice constants but different band gap energies. A schematic diagram of the material structure and the band diagram of a typical MQW structure is shown in Figure 3.17. The interesting quantum effect arises from the confinement of excitons in the narrow QWs of the material. Confining these excitons, which normally have a radius of about 300 Å, produces two effects. The first is that the exciton absorption line, which normally broadens into the band gap absorption with increasing temperature, remains resolvable even at room temperature. This effect is illustrated in Figure 3.18. The second effect is an induced anisotropy in the effect of an applied electrical field. The exciton absorption line is shifted by an electric field applied perpendicular to the structure layers. This is due to the QW’s enhancement of the exciton binding and consequent avoidance of the exciton dissociation until the higher than normal field strengths. These effects are illustrated in the Figure 3.19.

3.11.1 Description of the Quantum Well

The QW is formed by the discontinuities in the energy gap between the GaAs and the Al_xGa_1−xAs layers. The energy gap difference is described by

ΔEg(eV)=1.425x-0.9x2+1.1x3

where x is the mole fraction of aluminum. This energy gap difference is divided between the hole QWs and the electron QWs. Two theories exist regarding the magnitude of this QW split. The first is an 85:15 split, which means that 85% of the energy gap difference is at the conduction band discontinuity and 15% is at the valence band discontinuity. Another theory suggests a 57:43 split. It is interesting to note that the use of either one of these theories results in little or no difference in the magnitude of the energy shift. Figure 3.17 shows a schematic diagram of a QW structure indicating the band discontinuities.

To describe the exciton’s energy levels in a QW, we must first know the Hamiltonian. The Hamiltonian for this problem can be written as the sum of terms due to the contribution from the individual electron and hole, and the interaction between the two particles.

H=He+Hh+Hex⁢(3.15)

where

He=HKEze+Ve(ze)-eF⊥zeHh=HKEzh+Vh(zh)-eF⊥zhHex=HKEreh+Ve-h(r,ze/zh)

The z direction is defined perpendicular to the MQWS layers and z_e and z_h are the positions of the electron and hole, while r is the relative position of the electron and hole. The kinetic energy operators in the z direction are defined by

HKEze=-ℏ22me⊥x4d2dze2HKEzh=-ℏ22mh⊥zkd2dzh2⁢(3.16)

where

me⊥xk and mh⊥zk are the effective masses of the electron and hole in the z direction

V_e(z_e) and V_h(z_h) are the depths of the electron and hole QW potentials

F_⊥ is the electric field perpendicular to the MQWS layers

The effective masses for both of the discontinuity split ratios can be calculated as shown in Figure 3.20.

The kinetic energy operator for the interacting particles or exciton and their reduced effective mass are

HKEreh=-ℏ22μd2dr2μ=me‖*mh‖*me‖*+mh‖*⁢(3.17)

with me||* and mh||* being the effective masses of the electron and hole in the plane parallel to the layers. There are the normal bulk effective masses. The Coulomb potential energy due to the interaction of the electron and hole is

Ve-h(r,ze/zh)=-e2ε(|ze-zh|2+r2)1/2⁢(3.18)

3.11.2 Solution of the Exciton Energy

The next step to finding the energy levels is to compute the expectation value of H. This gives us the energy of the exciton

Eex=〈Ψ|H|Ψ〉⁢(3.19)

or

Eex=Ee+Eh+Eb

where

Ee=〈Ψ|He|Ψ〉Eh=〈Ψ|Hh|Ψ〉EE=〈Ψ|Hex|Ψ〉

E_e and E_h are the energies of the electron and hole respectively, which can be found using Equation 3.15. E_B is the exciton-binding energy which is found using Equation 3.15.

The wave function Ψ is written in a separable form as

Ψ=ψe(ze)ψh(zh)∅e-h(r)⁢(3.20)

where

∅e-h(r)=(2π)1/21λe[-r/λ]

The parameter λ is used as a variational parameter describing the amplitude radius of the exciton.

3.11.3 Determination of E_e and E_h

The problem of determining the energies of the electron and the hole in the QW is subject to a great deal of simplification. Initially it may be noted that, due to the separable nature of the wave function Ψ, Equations 3.19 reduce to the following:

Ee=〈ψe|He|ψe〉Eh=〈ψh|Hh|ψh〉⁢(3.21)

The next simplification is that a wave function describing an electron and hole in an infinite QW can be substituted for the wave function ψ_e or ψ_h if an effective well width L_x is used in place of the actual QW width. The longer well width effectively accounts for the fact that there is significant penetration of the wave function into the QW barriers.

The Schrödinger equation for the infinite well barrier in a uniform perpendicular electric field is

-ℏ22m⊥*d2dz2ζ(z)-(W+eF1z)ζ(z)=0⁢(3.22)

where ς(z) is either the electron or hole wave function. Note that the parameters m⊥* and e and the dimensions z and L_x all depend on the particle being investigated.

By making the substitution

Z=[2m⊥*(eℏF⊥)2]1/3(W+eF1z)⁢(3.23)

we can transform Equation 3.22 into an Airy differential equation:

ddZ2ζ(z)-zζ(z)=0⁢(3.24)

The solutions have the form

ζ(z)=bAi(Z)+cBi(Z)⁢(3.25)

where b and c are constants and Ai(Z) and Bi(Z) are Airy functions defined by

Ai(Z)=c1f(Z)-c2g(Z)Bi(Z)=3[c1f(Z)-c2g(Z)]

Further definitions are

f(Z)=1+∑n=1∞Z3n(3n)(3n-1)(3n-3)(3n-4)...3⋅2g(Z)=Z+∑n=1∞Z3n+1(3n+1)(3n)(3n-2)(3n-3)...4⋅3

while c₁ = 0.35503 and c₂ = 0.25882.

It is convenient to work with dimensionless parameters for the energy and the field. These are defined as

The energy W₁ is taken to be the lowest energy level of the particle with zero field. This is simply

W1=ℏ22m⊥*[πLX]2⁢(3.26)

The value W₁ is also used to define the unit for the electric field

F1=W2eLX

The values of Z at the QW walls are

Z±=Z(±12LX)

and therefore, using Equations 3.23 and 3.26 Z_± becomes

z±=[-πf]1/3(W+12f)⁢(3.27)

If we use the boundary conditions that ς(Z)_± = 0 at the QW walls we find from Equation 3.25 a resulting equation which completely determines the eigen energy:

0=Ai(Z+)Bi(Z-)-Ai(Z-)Bi(Z+)

This equation is completely independent of well width and effective mass.

3.11.4 Determination of E_B

The exciton-binding energy E_B can be rewritten in the following form

EB=EKEr+EpEr⁢(3.28)

E_KEr is the kinetic energy of the relative electron–hole motion in the layer plane. It is described by the following equation:

EKEr=〈∅e-h(r)|HKEreh|∅e-h(r)〉=ℏ22μλ2⁢(3.29)

E_pEr is the Coulomb potential energy of the electron–hole relative motion.

EpEr=〈Ψ|Ve-h|Ψ〉

In this case, we must use variational wave functions for ψ_e and ψ_h in Equation 3.20 [43]. These are written as

ψe,h=N(β)cos[πzLX]exp[-β[ZLX+12]]⁢(3.30)

Note that β (β_e or β_h), L_x (L_e or L_h) and z (z_e or z_h) all depend on whether we are using ψ_e or ψ_h. N(β) is a normalization function and is defined such that

N2(β)=4ββ2+n2Lx[1−exp(−2β)]β2

β is a variational parameter and is calculated by minimizing the function E(β) with respect to β²

E(β)=E1(0)[1+β24π2+χ[1β+2β4π2+β2-12coth(β2)]]⁢(3.31)

where the ground state energy at zero field is

E1(0)=ℏ2π22m⊥*LX2

and the dimensionless electro-static energy is

χ=|e|F⊥LXE1(0)

This is calculated separately for both the electron and hole.

Using Equations 3.18, 3.20, and 3.30 Equation 3.20 becomes

EpEr=-2e2πελ2N2(βe)N2(βh)∫θ=02π∫r=0∞∫ze=−Le/2Le/2∫2h=−Lh/2+Lh/2cos2πzeLeexp[-2βe[ZeLe+12]]

cos2πzhLhexp[-2βh[ZhLh+12]]Xrexp(-2r/λ)[(ze-zh)2+r2]1/2dθdrdzedzh⁢(3.32)

The integral over θ is trivial and the integrals over z_e and z_h must be done using numerical methods. We can calculate the integral over r using the following equation:

G(t)=2λ∫r=0∞rexp(2rλ)drt2+r2=2|t|λ[π2[H1(2|t|λ)-N1(2|t|λ)-1]]⁢(3.33)

where

H₁(u) is the first-order Struve function

N₁(s) is the first-order Neumann function or Bessel function of the second kind

The Struve function is defined by [44]

H1(u)=2π[u212⋅3-u412⋅32⋅5+u612⋅32⋅52⋅7-...]

while the Neumann function is defined by [45]

N1(s)=π2πY1(s)+(ln2-ϕ)J1(s)

where φ = Euler’s constant,

Y1(s)=-2sπ+2πln(s2)J1(s)-s2π[∑k=0∞{ξ(k+1)+ξ(k+2)}(-s2/4)kk!(1+k)]⁢(3.34)

J1(s)=(s/2)∑k=0∞(-s2/4)kk!(1+k)ξ(1)=-ϕξ(n)=-ϕ+∑k=1n-1k-1 n≥2

In summary, to calculate E_b for a particular filed we must first deduce β_e and β_h for the field, using a variational calculation, and then calculate E_KEr and E_PEr, adjusting λ variationally to minimize E_B.

3.11.5 Effective Well Width Calculations and Computer Simulation

In order to perform all the previous calculations, we must substitute an effective QW width for the actual well width. Note that for an infinite well the wave function has a value of zero outside the well walls. Because the wells in the MQWS are finite, the effective well widths for the infinite well model must be greater than the actual well widths to account for the significant penetration of the wave function into the finite QW barriers. There are two basic ways to find the effective well widths. One way is to try to match the wave functions of the finite well as closely as possible with the infinite well. Another way is to equate the energy of the first energy level in the finite well with the infinite well.

3.11.6 Summary of Calculation Procedure

With the full theory developed in the previous sections, a brief summary is helpful to explain the steps needed to obtain the exciton energy. Equation 3.19 shows that the exciton energy is broken down into three energy values. The quantum well is fully specified by what may be considered as three “fundamental” values. The aluminum concentrations of the barrier and the well material determine the band gap difference. The other parameter is the actual width of the well.

Step A—The effective masses of the particles, the conduction–valence band split, and the effective well widths are found as follows:

A1: The energy gap difference is used to determine the depths of the conduction and the valence band wells. There are the values V_e and V_h. A relatively arbitrary choice is made concerning the ratio of the conductance-valence band split.

A2: Figure 3.20 is used to determine the effective masses of the electrons, and light and heavy holes.

A3: Using the values determined in steps A1 and A2, along with the actual width of the wells, the effective well widths are determined by the methods previously described.

Step B—Values of E_e and E_h are found using the following steps.

B1: A “universal” curve is calculated using Equation 3.27. It is convenient to keep these values as a look-up table since the solution requires an iterative technique. This curve expresses the normalized energy w as a function of the normalized filed f through Equation 3.27.

B2: With a knowledge of w( f ) the actual energy of the particle W (which is actually E_e or E_h) is found by use of the Equation 3.26 to produce W1, F1, f, w, and W.

Step C—Finally, the binding energy E_B is determined. For each value of electric field F_⊥:

C1: Determine β_e and β_h for Equation 3.30. This is done by minimizing Equation 3.31 with respect to β.

C2: Using a value of λ and the above values of N²(β) calculate the value of E_PEr by performing the double integration over z_e and z_h in Equation 3.32.

C3: Calculate E_EEr using Equation 3.29.

C4: Repeat steps C2 and C3, adjusting λ to minimize E_B.

C5: Start at step C1 for the next field value.

Again, it is useful to use look-up tables for parts of Step C—in particular for the values of Equation 3.33.

3.11.7 Example: Fabrication of MQW Oscillator

An MQW could be fabricated from GaAs/AlGaAs heterostructure such as that shown in Figure 3.25. This structure would require about five mask and fabrication steps. The thin optical window, with no field applied perpendicular to the QW junction layers, will be transparent to a wavelength—denoted λ_low—of slightly lower energy than the absorption peak. Application of a filed will shift the absorption peak to lower energies and the window will become opaque to the incident wavelength λ_low. This operation, depicted in Figure 3.26 [41] shows the transmission of the device as a function of the applied bias. The proposed vertical material structure is shown in Figure 3.27 [46]. Figure 3.28 shows a photoluminescence scan of the material taken at room temperature (~300 K). Various important features and their wavelengths and corresponding photon energies are indicated in the figure. Most notable are the obviously well-defined exciton emissions, which would not be observable at room temperature without the QW structures. Peak 1 indicates the exciton from the equal width QW structure, denoted period B in figure. Peak 2 shows the emission from the unequal, period A wells. Peak 3 indicates emission from the wide band gap AlGaAs p-type cap layer (layer 8 in Figure 3.27). The energy of emission corresponds to an aluminum concentration of approximately 32%. Figure 3.29 shows another photoluminescence scan performed on the sample at a temperature of 12 K. Two effects are noted. First is the shift of the peaks. The second is the sharpening of the peaks due to the decrease of the phonon energy broadening.

3.12.1 History of the Magnetic Crosstie Memory

The crosstie magnetic effect in thin film magnetic material was discovered in 1958 by E.E. Huber. The magnetic thin film exists in two possible magnetic states: the Neel wall state which represents the fundamental magnetic domain wall; and the crosstie/Bloch line pair on the Neel wall, which represents the second magnetic domain state. The crosstie wall constitutes a transition form between Bloch walls of very thick films and the Neel walls of very thin films. As the film thickness is decreased, the Neel wall forms segments of opposite polarity, which reduces the magnetostatic energy in the wall. These segments are separated by perpendicular magnetization circles called Bloch lines, which minimize energy by forming spiked walls or crossties on alternate Bloch line locations. The crosstie Bloch line pairs represent digital information stored in the computer memory.

Work began in the early 1970s to develop a computer memory utilizing serrated strips of thin film 80–20 Ni–Fe permalloy data tracks that stored information in a series of crosstie Bloch line patterns on a Neel wall. The concept was similar to the serial propagation of bubbles in a magnetic bubble memory, but promised faster performance. The serial crosstie memory presented numerous development problems, however, illustrating the need for an alternative approach [48]. The random access approach was successfully demonstrated in early 1982 by Mentzer, Schwee et al. [49]. Its unique characteristics as a faster, nonvolatile, radiation- and temperature-insensitive, random access device make the crosstie memory a viable alternative for many applications. More recently [50], this technology is being developed for alternative flash memory, fast-start computer systems, biomedical storage devices, and dynamic RAM replacement. Significant efforts are underway as of this writing at IBM, Infineon, Naval Research Laboratory, and elsewhere.

3.12.2 Fabrication and Operation of the Crosstie Memory

Thin film permalloy (81–19 Ni–Fe) patterns on a silicon substrate support stable magnetization domain states which can be switched rapidly and detected nondestructively. The geometry of the basic memory cell is shown in Figure 3.30 [51]. The contours represent the direction of magnetization in the memory cell. The cell on the left represents a digital zero (Neel wall); and the cell on the right represents a digital one state (crosstie Bloch pair on Neel wall). The crosstie wall constitutes a transition state between Bloch walls of thick films and Neel walls of very thin films. The fundamental cell geometry supports two states—the Neel wall and the Bloch line-crosstie pair on the Neel wall.

Vacuum deposition of the permalloy alloy is performed in a magnetic field resulting in approximately 400 Å isotropic films with H_k < 10 Oe. Silicon substrate temperature and alloy stoichiometry are such that the material displays zero magnetostriction and maximum magnetoresistance ΔR/R > 2%. This ratio indicates the differential resistance between fields applied parallel and perpendicular to permalloy current flow, divided by the permalloy resistance. The current method of detecting the memory state utilizes this effect.

Switching between digital information states is accomplished through application of coincident in-plane fields (~10 Oe) produced by 5–7 mA currents flowing through overlying row and column conductors. The magnetic film changes states when a sufficient current is placed in the local area. Therefore, the column conductors zig-zag or meander in order to align the local electromagnetic fields of the column conductor with the row conductor in the local area of the memory cell. The test chip described by Mentzer et al. [52] (Figure 3.31) consisted of four 64-bit columns, a fifth reference column, column meander columns, column meander conductors, and four row conductors, providing access to 16 test bits. Five vacuum depositions and five photolithographic steps are required to fabricate the device. The permalloy memory cells are followed by a sputtered silicon nitride dielectric layer with vias for contacts to the permalloy, metallization for row conductors and contacts to the permalloy, another dielectric layer, and metallization for column conductors.

The electronic method for nondestructive read-out (NDRO) is performed by applying a positive field less than the annihilation level using the appropriate row conductor. The permalloy thin film has current applied. Therefore, when the current in the row conductor passes through a memory cell location in a crosstie state, a differential voltage is created. The differential detection voltage for a single bit obtained when a field is applied in the annihilation direction but less than the annihilation threshold is less than 300 μv. This low-detection signal level represents a significant challenge to the widespread implementation of the crosstie memory. As a result, Mentzer et al. [53] devised a signal extraction and processing architecture (Figure 3.32), which improved the signal level but increased the complexity of the device significantly.

The Mentzer/Lampe technique utilizes double-correlated sampling to remove offset nonuniformity fixed patterns by means of serial subtraction. Postdetection integration is implemented where signals accumulate constructively and directly while temporal noise is uncorrelated and adds only in terms of power. Integration of four such samples effectively doubles the signal to noise ratio. This approach also severely reduces the achievable memory density and increases power consumption, since the dominant factor dictating both density and power consumption is the size of the current drivers required to supply current through the permalloy columns. Clearly a technique is needed whereby the detection scheme is improved, in favor of a more direct readout—but still utilizing all of the useful features of the memory device (electronic write, radiation hardness, etc.).

3.12.3 Potential Optical Detection Scheme

Optical detection could supplant the magnetoresistance detection approach, which limits access times to approximately 500 ns. This is too slow for many primary memory applications. The optical technique will afford read-write cycle times that are competitive with the fastest available technologies on the market. In fact, the crosstie memory access times may prove to be the fastest achievable, with the optical addressing scheme described herein.

The detection circuit for the optical detection (read) scheme requires simpler circuitry; so there will be not only performance gain but also significant improvement in cost and producibility. Additional limitations of the magnetoresistance effect include the requirement to heat the substrate to increase crystallite size and to provide the necessary magnetoresistance ratios, along with temperature limitations. These limitations are eliminated with optical access and detection using the Faraday magneto-optic effect.

Magneto-optic Faraday rotation occurs when linearly polarized light is transmitted through a material parallel to the direction of a magnetic field in the material. The amount of rotation φ is proportional to the magnetic field M and the path length t, such that φ = VHt cos θ, where V is the Verdet constant (rotation per unit length per unit magnetic field), and θ is the angle between the magnetic field and direction of light propagation. In the proposed crosstie memory detection configuration, the presence or absence of a crosstie wall is used to produce a change in the transmitted optical power density through crossed polarizers that are rotated from extinction by an angle equal to the amount of Faraday rotation produced. Differences in transmitted light intensity through different memory locations thus provide detection of stored information. The Neel wall state does not have a magnetic field perpendicular to the permalloy surface while the crosstie state does.

Figure 3.33 shows the permalloy pattern to be used to write information on the memory device. Creation of crossties is performed by the application of coincident currents flowing through conductor strips insulated by nitride layers. Windows should be added to the nitride and conductor layers to define the spot size of the transmitted light for optical detection of the crossties. Polarizers can be either grown or bonded to the device as wafer processing permits. Laser radiation will be applied through a particular memory block or array accessed via electro-optic switching.

Detection of transmitted light intensity is accomplished with a silicon PD array. Random access or parallel addressed read may also be accomplished through the detector array, rather than electro-optic switching. Current levels produced by changing light intensities resulting from partial extinction of Faraday rotated light are amplified and coupled to a comparator with sensitivity controls allowing identification of logic levels. The device configuration is illustrated at Figure 3.34.

Assuming a crosstie magnetic field equal to the saturation magnetization, we have a specific Faraday rotation in permalloy equal to 1–2 × 10⁵ deg cm⁻¹. This provides a rotation of approximately 2° with a permalloy thickness of 600 Å, at λ₀ = 830 nm. Thin film and multilayer dielectric polarizers supply a 3.5% change in transmitted intensity per degree of analyzer rotation with the polarizer at 45° from extinction with respect to the input polarization. A fixed insertion loss of 15% for the polarizers, and a loss due to absorption in the permalloy of 2.7% for 600 Å thickness (α = 6 × 10⁵ cm⁻¹) along with a 50% loss due to initial polarizer rotation, provide a decrease in optical power density form polarizer to detector of 65%. This level is decreased by 3.5% polarizer transmittance per degree of rotation X2 = 7% due to the presence of a crosstie. If we couple 3 mW of power into the system, the PDs will see changes of 74 μW on a background of 1050 μW.

PDs can produce 25 mV μW⁻¹ μs⁻¹ integration time at 12 μm center-to-center spacings. A power level of 1050 μW then provides a 650 mV signal using 25 ns integration time, with a 46 mV change in voltage for crosstie detection. Excellent signal to noise ratios are thus possible for low false alarm probabilities. The signal may be compared to a reference zero location electro-optically switched simultaneously, and differentially amplified for further signal processing. Such a system provides excellent high-speed detection for the crosstie memory, and eliminates the problems associated with electrically addressed readout.

3.12.4 Radiation Hardening Considerations

Radiation environments in which such a memory may be required to survive are a strong function of the particular system deployment. Tactical man-operated systems are required to survive only modest ionizing dose rates, low accumulated doses and low neutron levels, with man being the limiting factor. Strategic ground-based systems are required to withstand sharply higher levels. Space-based systems with no nuclear weapon environment must survive only the natural Van Allen belt energetic electron and proton environments, which may range up to several tens of kilorads (Si0) depending on the orbit and the duration of the mission. Military systems in space must survive the nuclear weapon environments of high ionizing dose rates, high accumulated doses and neutron fluences. The system memory requirements in terms of organization, power, speed, etc., will all be different in each of these applications; but in general will tend toward higher speed, low power, high density, nonvolatility, and low soft error rates.

The optical read crosstie memory discussed affords all these features in addition to nuclear hardness. Since the optically read memory is inherently free from soft errors, there will be no need to increase the word length to allow for error detection and correction. Normally, 5 bits of a 16 bit word are utilized for double error, single error correction; so an immediate savings of 30% of the memory size and power supply requirement is realized. The radiation environments in which memories may be required to survive vary with application, but may include ionizing dose rates in excess of 10¹² rad (Si)/s, total accumulated doses of up to 10⁷ rad (Si), and neutron fluences of up to 10¹⁴ n cm⁻².