There are tremendous differences between the laser and the transistor, but it is possible to draw their attributes together by building a transistor laser. This novel device that produces its electrical and optical outputs simultaneously promises to revolutionize data transfer, enabling new architectures capable of operating at incredibly high bit rates, says Milton Feng from the University of Illinois, Urbana-Champaign.

The semiconductor industry rakes in billions and dollars from the manufacture of devices invented way back in the middle of the twentieth century. With the benefit of hindsight it is clear that the most important of them all is the transistor, a device invented by John Bardeen and Walter Brattain in 1947, which has been the key building block in the development of microelectronics, integrated circuits, consumer electronics, and the computer industry. Not far behind the transistor is the visible LED, made for the first time by Nick Holonyak in 1962, and the laser diode, independently invented in that same year by Holonyak and Robert Hall. These two optoelectronic devices have provided a great foundation for revolutionizing display, lighting, and information technology.

Although the performance of all these devices has come on in leaps and bounds over the intervening decades, none can simultaneously deliver an electrical signal and a laser output. The invention of such device had to wait until 2004, when I, Milton Feng, in partnership with co-worker Holonyak, produced the world’s first transistor laser.

This revolutionary semiconductor device that offers threeport operation – an electrical input, an electrical output and an optical laser output (see Figure 1) – has the potential to a make important contributions to integrated photonic and electronic integrated circuits, the computer industry and new information technology. Amongst these many promises, it is capable of redefining the approach made to the transfer of digital data. Today, PCs operate solely in the electrical domain, and hooking up to the internet requires an infrastructure involving transmitters and receivers that can provide an interface with the optical domain. The transistor laser, however, is capable of performing all these functions itself. One of its roles could be to act as an optical interconnect that could allow incredibly fast data flow to and from memory chips, graphics processors and microprocessors.



Fig. 1 Three-port operation of the transistor laser provides an electrical input (port 1), an electrical output (port 2) and an optical laser output (port 3).

Our transistor laser, which emits infrared light, is a modified, high-speed HBT with a quantum well in its base region. In conventional high-speed HBTs, which inevitably operate at a high current density, the base provides the pathway for electrons to travel from the emitter to the collector. In our device, the quantum well in the heavily-doped base traps some of these electrons, which diminishes the transistors gain, but allows this device to realize radiative recombination between holes and electrons. Thanks to the geometry of our device – the chip has cleaved facets that act as mirrors – the light that is emitted is bounded by a cavity, enabling stimulated emission, one hallmarks of a laser.

Our first transistor laser needed to be cooled with liquid nitrogen. But a year later it could be run at room temperatures, thanks to improvements in MOCVD growth and the design of the quantum well. Since then we have focused on improving the quality of light output from our transistor laser and understanding its electrical behavior.

Revolutionary modulation speeds

One of the really encouraging attributes of our transistorlaser is its incredibly fast radiative recombination lifetime:it is below 30 ps. This can spur the direct-modulationbandwidth in an LED to 10 GHz, and to 100 GHz in alaser. The far higher bandwidth will accelerate thedeployment of LEDs and lasers in optoelectronicinterconnects and open the door to a new generation ofhigh performance, electronic-photonic integrated circuits.

Traditional laser diodes suffer from a resonance peak in the frequency response. To combat this, resonance compensation circuits are included in transistor laser driver circuits. Our transistor lasers, however, do not have to contend with this thanks to a shift in the carrier-photon damping ratio and elimination of the resoance peak. Thanks to these attributes, our laser transistors could be used to build an ultra-low power transmitter and array for 100 Gbit/s Ethernet and optical interconnect applications. In addition to the high speeds, there is also the possibility to tap into our device’s non linear characteristics, and exploit flexible signal mixing and processing.

An additional weakness of the conventional laser is a pulsation or “ripple” in its output. This phenomenon is well understood. It was observed in 1959 in masers under certain pump conditions, and has been studied in detail by the researchers Statz and deMars. They explained its occurrence in the 1960s by studying the transient solution of a pair of coupled carrier-photon rate equations describing the dynamical interaction between population inversion and cavity electromagnetic energy.

The unwanted self-resonance seen is these masers also plagues today’s laser diodes used for data communication. Here it causes a hike in the bit error rate, which is countered with expensive, complex peripheral circuits. Typically passive low-pass filters, such as Bessel filters with a fixed cut-off frequency, are employed to “filter out” the resonance frequencies. But this addition comes with a big performance penalty: it restricts the laser’s transmission bandwidth to below its resonant frequency.

One of the great strengths of our novel, three-port device is that it can produce resonance-free semiconductor laser behavior. This stems from the incredibly fast base spontaneous recombination lifetime, which is typically just 29 ps. To realize this we use an approach that would fail in today’s workhorse for data communication, the p-i-n double heterojunction laser. This involves building a structure that tilts the injected carriers and diffuses them across a thin, oppositely doped quantum well base active region. Slowly recombining carriers are removed, and “fast” recombining carriers favored (see Figure 2). It follows that it is possible that the intrinsic spontaneous recombination time in the base of the transistor can be “clamped” at the same order of magnitude as the quantum well base region transit time, typically 10 ps.



Figure 2. (a, top) Bandwidth (20 GHz) (b, above) Eye diagram (13.5 Gbit/s)

We have studied the behavior of our transistor laser in more detail by considering its small-signal linear optical response. A damping factor is included in our calculations. One insight gained from this effort is there are at least three approaches to reducing the resonance peak: speeding the base spontaneous recombination lifetime; increasing the natural resonance of the system; and reducing the ratio between the base current and the threshold current. Analysis also reveals that it is possible to realize a “critically damped” condition that eliminates carrier-photon resonance.

The small-signal linear optical response of our device has been calculated for spontaneous recombination lifetimes of 2, 10, 50 and 250 ps. For these calculations we have assumed a photon lifetime of 2.5 ps and a value of five for the ratio of base current to threshold current.

In addition, we have measured and fitted an optical frequency response to our transistor laser. This highlights the absence of carrier-photon resonance, which results from the “fast” base spontaneous recombination lifetime. In agreement with our model, there is a slight resonance in the output of our laser, which is less than 3 dB and only seen at higher bias. What is pleasing is that a resonancefree response of the tilted-charge transistor laser is observed at a range of biases: (a) IB = 30 mA; (b) 40 mA; (c) 60 mA; and (d) 100 mA.

It is worth noting that the Statz and deMars coupled carrier-photon equations do not include parasitic charging delays. To cater for this, we have developed a physically based transistor model, which includes parasitic charging delays to predict microwave frequency response and digital eye-diagrams.

A consequence of the multi-port capability of the transistor laser is the need to re-formulate Kirchoff’s law, which is widely used to analyze and design conventional circuits. In order to cater for our transistor laser, this law must include energy conservation, and not simply current and charge. We have recently done just this, and published a paper detailing these efforts in the Journal of Applied Physics. Our novel transistor laser clearly holds great promise. It is still early days, but what is clear is that this multi-port structure offers a vast improvement in topological and device-to-device system design freedom. Thanks in part to these attributes, it promises to offer a leap in the performance of electrical-optical integrated circuits that is impossible to conceive with either the transistor, or even more limited two-terminal diode.

This work is sponsored by DARPA and ARO