Slashdot Mirror
Experimental Transistor Breaks 600 Gigahertz
neutron_p writes "The goal of a terahertz transistor for high-speed computing and communications applications could now be within reach. A new type of transistor structure, invented by scientists at the University of Illinois, has broken the 600 gigahertz speed barrier.
A new type of transistor - built from indium phosphide and indium gallium arsenide - is designed with a compositionally graded collector, base and emitter to reduce transit time and improve current density. With their pseudomorphic heterojunction bipolar transistor, the researchers have demonstrated a speed of 604 gigahertz - the fastest transistor operation to date."
More and more we here about these new HBT circuits that are faster than all get out.
The truth is that nothing will replace CMOS anytime soon. The infrastructure is already there, and it is being optimized over and over again and has a huge work force to man it.
I once heard someone ask Intel is they ever plan to switch to HBT for speed. Their response is, and will probably be for a while, that why would they switch technologies after investing $50 billion a year in their CMOS foundries etc.
These advancements may never make it to the point that the average consumer will take notice of them.
And it may be that these academic inventions will never find any market relevance.
It all depends on the wiring delay and how many transistors deep a pipeline stage is.
fMax of a pipeline stage is 1/(switching times+wiring delays) under worst case thermal conditions. The wiring delays will stay about the same unless they're also improved by the new process, which is unlikely.
A 600GHz transistor, with really deep pipelines like the P4, and very good interconnect technology might allow 20-50GHz operation; but there are many other things to contend with (like thermals/dissipation) that can limit speed. Thermals, in turn, depend on the amount of capacitance being switched, which isn't specified here.
I think these transistors, if found to be manufacturable, will probably be used in communications not digital logic.
Indeed. The transistors used for digital circuits (i.e., computers) are mostly MOSFETs. The chief benefit of MOS transistors is that no current goes into the gate, so power is only used when switching from one state to the other (i.e. from a 1 to a 0).
Bipolar transistors have a base current (albeit small), so they draw power even when responding to a constant signal. However, they're faster and can output a lot more current than MOSFETs, so they do have plenty of other applications.
I know special methods exist to predict the f_s from low-frequency measurements. Maybe they measure the amplification at a some 'low' frequencies (GHz range) and extrapolate the gain-bandwidth pruduct from this?
Spectrum analyzers could "see" up to 325 GHz directly in the early Eighties. So I'd guess that newer and better waveguide mixers are available now. A Tek 2782 or 2784 analyzer could theoretically display a harmonically-downmixed signal 1.2 THz, although I have no idea how you were supposed to acquire the signal in the first place.
You may not be able to see a single one-picosecond pulse in the time domain, but if you fire off a bunch of them in succession, you can build a picture of the waveform with repetitive sampling techniques. Technology was available in the 1960s to perform repetitive sampling in the 20-picosecond regime, so someone like Tek or Agilent or Picosecond Pulse Labs may have a sampling gate that can do the job.
I would recommend surfing around at PPL's site if you're seriously interested in this stuff. There may also be some photonic tech involved in the measurement; I haven't RTFA yet.
Dahlmann tightly grips the knife, which he may have no idea how to use, and steps out into the plain.
By heterodyning with (multiplying by) a lower frequency. Look up formula for sin(at) x sin(bt).
Note also that harmonics of a given frequency can be created by passing it through a nonlinearity.
Contribute to civilization: ari.aynrand.org/donate
Filters of known frequency response can be made by knowing only their geometry. Pass the signal through several filters of different frequency responses (one at a time) and feed the output of the filter into a resistive material. Measure the temperature of the resistive material. The peak frequency of the filter which warms the resistive material the most is the (approximate) frequency being generated.
Contribute to civilization: ari.aynrand.org/donate