A Semiconductor-Based Photonic Memory Cell

See allHide authors and affiliations

Science  26 Feb 1999:
Vol. 283, Issue 5406, pp. 1292-1295
DOI: 10.1126/science.283.5406.1292


Photonic signals were efficiently stored in a semiconductor-based memory cell. The incident photons were converted to electron-hole pairs that were locally stored in a quantum well that was laterally modulated by a field-effect tunable electrostatic superlattice. At large superlattice potential amplitudes, these pairs were stored for a time that was at least five orders of magnitude longer than their natural lifetime. At an arbitrarily chosen time, they were released in a short and intense flash of incoherent light, which was triggered by flattening the superlattice amplitude.

Storing light for appreciable amounts of time is not an easy task because light is always propagating. In principle, light can only be stored directly by guiding its path in a loop back to the origin [for example, in a cavity in which a light beam is folded back and forth between two mirrors or a whispering gallery–type resonator (1)] or by guiding the light along a coil of optical fiber (2). However, to reach delay times of the order of 1 μs, one must use mirrors that lose very little light to obtain the necessary 300 reflections in a 1-m-long cavity, or equivalently, one must use 300 m of fiber to form the loop. Another recently realized approach to light storage uses Anderson localization (3); for this method, the light path is prolonged by passing it through a nonabsorbing but highly scattering medium. In addition to being bulky, these techniques in general do not allow for easily variable delay times.

For optical signal processing (2) and pattern recognition, however, one would like to store the photonic signals in an array of pixels that can be implemented on a chip. Analogous to an electronic memory where electrons are stored in a capacitor, each pixel could act as an optical memory for intermediate storage or delay of selected optical bits. Such memory cells are the most important part that is still missing from an optical network unit (2). Ideally, such a device would be a small container in which an incoming optical signal could be stored for an arbitrarily chosen time and then be released again as light. The switching speed should be of the order of 1 gigabit/s for local area network applications. Attempts to develop this device rely on a hybrid solution, in which a “smart pixel” (4) registers the light with a conventional detector, converting it to an electrical signal that is then stored in an electronic memory cell. At a given time, the signal can be reconverted to light by using, for example, a laser diode as an emitter. This approach, however, is a complex solution requiring many different techniques and components for a single pixel.

Here, we report the realization of a conceptually different, yet simple and potentially very efficient, storage cell for optical energy, which is based on a field-effect tunable lateral potential modulation in the plane of a semiconductor quantum well (QW). In such a cell, light energy can be locally accumulated and stored for many microseconds by converting it into spatially separated electron-hole (e-h+) pairs. At a chosen moment, the light can be reemitted from the cell in a short and intense flash. The radiative recombination lifetime of photogenerated e-h+ pairs can be voltage-tuned over many orders of magnitude from ∼1 ns to several tens of microseconds (much longer storage times are anticipated). Analogous to electronic dynamic random access memory (DRAM), an optical DRAM can thus be realized, which we envision to be potentially very attractive for optical pattern recognition and image processing. Deliberate release of the light is achieved by triggering the radiative recombination of the stored charges. Each pixel cell can be very small, and storage and emission are controlled by low-voltage signals. However, the photoluminescence (PL) mechanism exploited for intermediate storage does not conserve the coherence of light.

For optical storage, both large absorption and long storage times are desired. Direct band gap semiconductors have strong absorption above the band gap, but the radiative lifetime of the photogenerated e-h+ pairs usually is in the range of nanoseconds (5). In contrast, indirect band gap semiconductors have longer radiative lifetimes, yet the interband absorption is small. We combined strong interband absorption with variable lifetimes that were tailored by a low-voltage–induced lateral electrostatic superlattice. The lifetime of the radiative e-h+ recombination in a QW can be tuned by the overlap between the e and h+ wave functions (5). When this overlap is considerably reduced by spatially separating the photogenerated e-h+pairs, very long lifetimes can be achieved without losing the large absorption of the host material. Radiative recombination of the stored electrons and holes through a fast direct transition can then be triggered at a later arbitrary time by removing the electric potential modulation that is responsible for their spatial separation.

Recently, Rocke et al. (6) and Rufenacht et al. (7) showed that a spatial separation can be used to prolong the lifetimes of photogenerated electrons and holes in a QW. Rocke et al.(6) created a lateral potential modulation with a piezoactive sound wave propagating on the sample surface. At a remote location, radiative recombination of the transported bipolar charges was induced by flattening the potential modulation. Rufenacht et al. (7) used the built-in potential of a special coupled double QW and triggered the recombination by optically exciting the electrons into an energy state common to both QWs.

To store photonic signals, we used interdigitated gate electrodes on the surface of a QW heterostructure (Fig. 1A). A voltage difference that was applied to the gates resulted in a lateral type II potential superlattice with an electrically tunable amplitude in the plane of the QW (Fig. 1B). Upon illumination and for lateral electric field components that are strong enough to prevent the formation of excitons, the photogenerated electrons and holes drifted according to their respective charge along the gradient of the lateral potential into laterally opposite directions and accumulated in their respective potential minima (6). They can be stored in these minima for appreciable amounts of time because the overlap of the e and h+ wave functions is dramatically reduced, and they are confined to the potential minima in the plane of the QW. To release the stored light, we switched off the lateral potential modulation, which resulted in a flat band condition as in the unperturbed case. Attracted by their Coulomb interaction, the electrons and holes approached each other and hence increased the overlap of their wave functions dramatically. The result was an intense flash of luminescence light shortly after the potential modulation was switched off.

Figure 1

(A) The optical memory cell, consisting of a semiconductor QW with two interdigitated metal gate electrodes on the surface. (B) Schematic cross section of the storage cell and the lateral electrostatic potential modulation in the plane of the QW caused by the applied gate voltagesV 1 and V 2. Photogenerated e-h+ pairs were spatially separated and stored in the modulated potential as indicated. The positions of the conduction band edge (CB), the Fermi level (EF), the valence band edge (VB), and the dominant components of the electric field (E field) are also indicated.

This storage mechanism relies on the electrons and holes being separately confined in the plane of a QW and is, in principle, independent of the choice of semiconductor materials. In our experiment, we used a direct band gap GaAs-based QW that was grown by molecular beam epitaxy; the layered structure consisted of a Si-doped back electrode (donor density N D = 4 × 1018 cm–3) that was grown on an undoped GaAs buffer and a 20-nm-wide undoped QW that was sandwiched between two short-period AlAs-GaAs superlattice barriers. The QW and the back electrode were separated from the surface by distances ofd QW = 60 nm and d BE = 390 nm, respectively. On top of the sample surface, we deposited a 10-nm-thick, semitransparent interdigitated Ti gate electrode of periodicity a = 4000 nm. The width of the individual metal stripes was w = 1100 nm, and the total gate electrode covered an area of A = 200 μm by 200 μm. Both the total potential with respect to the back electrode and the amplitude of the lateral potential modulation could be controlled independently. In optimized samples, lateral modulations that are even larger than the effective band gap should be achievable. The resulting structure then would resemble a nipi (n-type/intrinsic/p-type/intrinsic) doping superlattice (8). In contrast, however, our superlattice was not created by a permanent doping modulation but rather by an easily tunable lateral potential modulation.

In the above geometry and with low voltages applied to the gates, we generated strong electric fields of up to 105 V/cm, containing both vertical and lateral components. The vertical electric field components were strongest in areas underneath the gate electrodes and influenced the optical properties of the QW through the quantum-confined Stark effect (QCSE) (9). The lateral electric field components that are strongest in areas between the gate electrodes were responsible for the spatial separation of the e and h+ states. For the comparatively large geometry and typical gate bias used, the wave function overlap between the spatially separated e and h+ states was essentially zero.

The sample, stored at temperature T ≈ 100 K was illuminated by a pulsed laser diode of pulse power P = 165 μW at a wavelength of λ = 820 nm (1.58 eV) (that is, below the band gap energy of the QW barrier material), resulting in the generation of e-h+ pairs in the QW. The light was focused on the gate with a spot size of ∼100 mm in diameter. The luminescence from the sample was analyzed with a monochromator and a gated photomultiplier. The PL spectra from the sample for different experimental conditions at various discrete times were plotted (Fig. 2, A through D) with the corresponding lateral potential modulation. The gate voltages were chosen so that the Fermi level remained in the effective band gap; that is, we did not have free carriers in either band without illumination.

Figure 2

Photoluminescence spectroscopy verified the storage of photonic signals in the optical memory cell. The graph displays spectra at different experimental conditions and times. Laser and gate voltages were switched as indicated on the time axis. (A) Without potential modulation (V 1= V 2 = 0 V; ΔV = 0 V), a single narrow PL line of photogenerated excitons was observed. (B) The application of small voltages (V 1 = –0.4 V and V 2 = +0.6 V; ΔV > 0 V;) spatially separated electrons and holes in the lateral potential superlattice as indicated. (C) When the laser was switched off, no luminescence could be detected, and electrons and holes remained stored. (D) By resetting the gate voltage modulation to zero (V 1 = V 2 = 0 V; ΔV = 0 V), a strong luminescence signal was observed from the recombination of the stored electrons and holes. au, arbitrary units.

During the first step of the experiment (Fig. 2A), no bias difference was applied to the gates (voltage V 1 =V 2 = 0 V; voltage difference ΔV = 0 V), and the sample was illuminated with the laser. The interdigitated gate behaved like a homogenous electrode, and the PL from the sample at an energy of about 1.5225 eV was the “usual” spatially direct PL line of a QW with a flat lateral band structure (10). At the given reduced sample temperature ofT = 100 K, this PL line was narrow and intense, which is typical for its excitonic origin. During the next time interval (Fig. 2B), a gate bias difference was applied to the gates (V 1 = –0.4 V and V 2 = +0.6 V; ΔV = 1 V) while the sample was illuminated with the laser. Under this condition, the lateral potential modulation resulted in a well-resolved splitting of the PL into two lines (one at 1.515 eV and a weaker line at 1.5227 eV). This splitting and the relative intensities of the lines are characteristic for the induced strong lateral potential modulation, which is caused by the local QCSE underneath the individual gate stripes (10–12). Their intensity ratio indicates exciton transport to areas underneath the stripes, where the QCSE is strongest and the excitonic energies are minimal. The strong overall decrease of the PL intensity was caused by the spatial separation of the e-h+ pairs (Fig. 2B). The time interval (Fig. 2B) during which the potential modulation and the laser were “on” simultaneously was referred to as the “loading time” t L because the minima of the lateral potential modulation were filled with electrons and holes.

The laser was then switched off (Fig. 2C), whereas the potential modulation was kept on. During this “storing time”t S, no PL was observed. The duration oft S can be chosen at will and was varied between typical durations of 1 μs (as in Fig. 2) and 35 μs. This period is ∼104 times longer than the radiative lifetime of excitons in an unmodulated QW. During t S, the photogenerated electrons and holes were spatially separated by about one- half of the modulation period and trapped in the lateral potential wells of the conduction and valence bands, respectively. PL was suppressed, as the wave function overlap in this case was negligible. The end of t S was defined by the instant at which the lateral potential modulation was turned off (V 1 = V 2 = 0 V; ΔV = 0 V). Because they were free to move under their mutual Coulomb attraction, the electrons and holes radiatively recombined, and a bright flash of luminescence light was observed coming from the sample (Fig. 2D), documenting the accumulation of electrons and holes during the loading interval (Fig. 2B) and their long-term storage (Fig. 2C). Because the energetic position of the luminescence flash (Fig. 2D) was identical with the position of the original PL lights in Fig. 2A, we conclude that the recombination of the stored charges was excitonic and originated from the spatially direct transition of a QW with homogeneous lateral bands, as defined by the choice of gate voltages. A proper choice of the total potential applied to the gates can be used to determine the wavelength of the delayed light (10, 12).

A time-resolved measurement of the intensity of the flash in Fig. 2D is shown in Fig. 3. The storage time in this case was t S = 17 μs. The duration of the flash was shorter than the limit of our time resolution (20 ns), but we suppose, however, that the recombination time was much faster according to its excitonic nature. Our experiment showed that low-intensity photonic signals can be accumulated into the structure over long times and then be released at once in a bright short flash. The intensity of this flash as a function of the storage time t Sis shown in Fig. 4. Even for the present unoptimized sample design, very long storage times can be achieved. We restricted ourselves to measurements of times up to 35 μs, where the detected signal was still well above the noise level, but we anticipate the feasibility of storage times that are >100 μs. The slight decrease of the flash intensity with increasing storage time is thought to be caused predominantly by nonradiative recombination of the stored charges, such as electron tunneling to the gate electrodes. Comparing the intensities of the direct PL (Fig. 2A) with the time-delayed luminescence of Fig. 2D already yields a 12% recovery of the signal for a storage time of 1 μs. The loading time in our experiments was typically t L = 3 μs at a pulse intensity ofP = 164 μW. For a longer t L, the detected flash intensity tended to saturate because eventually the stored electrons and holes screened the lateral potential modulation. We estimate that the maximum charge densities obtained in our device were of the order of 1011 per square centimeter.

Figure 3

Time-resolved luminescence signal after a storage time of t S = 17 μs. Resetting the gate voltage difference to zero (ΔV = 0 V) triggered an intense flash of light that left the cell within nanoseconds.

Figure 4

Time-delayed luminescence intensityI (I 0, intensity of the PL att S = 0) for different t Sreflecting storage times in excess of 100 μs.

The storage of photonic signals by the above means is not restricted to the energy range and material combination that we chose. Different material combinations and sample designs (for example, the use of microcavities) are waiting to be explored. The storage cell presented here works well at temperatures of 100 K. At higher temperatures, the excitons were thermally dissociated as the thermal energy exceeded the exciton binding energy. Then, the PL signals of our present sample became quite weak and spectrally broad. However, newly designed and optimized structures promise room-temperature operation in the near future.

  • * To whom correspondence should be addressed. E-mail: achim.wixforth{at}


View Abstract


Navigate This Article