Advanced Topics In Solid State Devices EE290B. Will a New Milli-Volt Switch Replace the Transistor for Digital Applications?

Advanced Topics In Solid State Devices EE290B Will a New Milli-Volt Switch Replace the Transistor for Digital Applications? August 28, 2007 Prof. Eli Yablonovitch Electrical Engineering & Computer Sciences Dept. University of California, Berkeley, CA 94720 1

Read the current going through a resistor, in the presence of noise: Required ( Δi) ( Δi) voltage 2 2 V = 2q i Δf...Shot Noise 4kT = Δf...Johnson Noise R = ir >> 2kT / q ~ 50mVolts Signal to Noise Ratio i Required power iv With a safety margin: Energy Consumed = > > ~ 2q Δf 2q 40 i 2q iδf Δf kt per = 2kT q bit i 2q Δf = 4kT Δf processed

What will be the energy cost, per bit processed? 1. Logic energy cost ~40kT per bit processed 2. Storage energy cost ~40kT per bit processed 3. Communications currently >100,000kT per bit processed. 3

There are many type of memory possible: 1. Flash 2. SRAM 3. Dram 4. Magnetic Spin 5. Nano-Electro-Chemical Cells 6. Nano-Electro-Mechanical NEMS 7. Moletronic 8. Chalcogenide glass (phase change) 9. Carbon Nanotubes Similarly there are many ways to do logic. But there are not many ways to communicate: 1. Microwaves (electrical) 2. Optical 4

What is the energy cost for electrical communication? V V 2 noise 2 noise R = = 4kT 4kT R Δf Δf Signal Energy Noise Power per bit = 4kT per bit All information processing costs ~ 40kT per bit. (for good Signal-to-Noise Ratio) Great! So what s the problem?

The natural voltage range for wired communication is rather low: V V V V V 2 noise 2 noise 2 noise 2 noise noise = 4kT R Δf 1 = 4kT R RC 1 = 4kT C 4kT q = q C = 4kT / q 123 100mVolts V 1 mvolt q { / C 10μVolts The natural voltage range for a thermally activated switch like transistors is >>kt/q, eg. ~ 40kT/q or about ~1Volt Voltage Matching Crisis at the nano-scale! If you ignore it the penalty will be (1Volt/1mVolt) 2 = 10 6 The wire wants 1000 electrons at 1mVolt each. (to fulfill the signal-to-noise requirement >1eV of energy) The thermally activated device wants at least one electron at ~1Volt.

The other Moore's Law, for energy per bit function 10 8 Critical Dimension 10μm 1μm 100nm 10nm Gates including wires Gates only Technology Gap 10 7 10 6 10 5 10 4 10 3 10 2 10 Energy per Bit function (kt) 1 Transistor Measurements by Robert Chau, Intel 1960 1980 2000 Year 0.1 2020 2040 2060 Shoorideh and Yablonovitch, UCLA 2006

p. 114 "In addition, power is needed primarily to drive the various lines and capacitances associated with the system. As long as a function is confined to a small area on a wafer, the amount of capacitance which must be driven is distinctly limited. In fact, shrinking dimensions on an integrated structure makes it possible to operate the structure at higher speed for the same power per unit area."

A low-voltage technology, or an impedance matching device, needs to be invented/discovered at the Nano-scale: transistor amplifier with steeper sub-threshold slope hν photo-diode ~1eV nano-transformer - + + + + MEM's switch V G + Cryo-Electronics kt/q~q/c giant magneto-resistance spintronics Cu solid electrolyte Cu Electro-Chemical Switch

in An amplifying transistor as a voltage matching device: Small voltage in Large voltage out out Amplification of weak signals has an energy cost! Amplification of weak signals has a speed penalty! ln{i} Current steeper sub-threshold slope Gate Voltage V g correlated electron motion?

Cryo-Electronics: kt q ~ q C RSFQ

Nano-Mechanical Switch: + + + - V G + I ~ exp(-3qv G /kt) for 3 charges on the MEM's tip After M. A. Baldo

giant magneto-resistance + "spintronics" These switches are made of metallic components and are of inherently low impedance

Transpinnor Structure: I signal Gate Current Gate Insulator I signal 6:1 Resistance Change in Tunnel Magnetoresistive (TMR) stack [1] Current Gate Source Drain Drain B Field Source Ru 15nm Ta 5nm Ferromagnetic CoFeB 3nm MgO 1.5nm Tunnel Barrier Ferromagnetic CoFeB 3nm Ru 0.8nm CoFe 2nm Antiferromagnetic MnIr 8nm NiFe 5nm Ta 5nm Ru 50nm Ta 5nm Si (001) Substrate Device Area 1μm 2 [1] Ikeda et. al., Japanese Journal of Applied Physics, Vol. 44, No 48, pp. L1442-L1445 B Field Magnetization

Complementary Transpinnor Logic +V +3mV 5μA input 2.275kΩ or 500Ω 5μA output 500Ω or 2.275kΩ Output Power = 1.6*10-8 W Total Power = 2.5*10-8 W Efficiency=65% -V -3mV Problems: On/Off ratio is only about 5:1 Still takes too many μamps to switch

A Photo-Diode for impedance matching: hν nhν ~1eV multi-photon millimeter-wave "optical" plasmonic wires Problems: Ten photons are needed to assure a good Bit Error Ratio. hν nhν Source efficiency may be only 10%. Energy penalty 100 40kT Miller DAB "Optics for Low-Energy Communication Inside Digital processors Quantum Detectors, Sources, and Modulators as Efficient Impedance Converters Optics Letters 14 (2): 146-148 JAN 15 1989

Nano-transformers are feasible using tapered plasmonic transmission lines: +++ - - - + + + - - - + + + - - - + + + - - - + + + - - - + + + ++ - - - - - Normal Impedance High Impedance

Electro-Chemically Driven Metallic Switch: 1nm Cu solid electrolyte Cu - V G +

The Esaki Diode as a Switch: The sub-threshold slope for tunneling depends on the steepness of the band-edges: E F Current band to band tunneling Sharp Step Diffusion current Bias Voltage

The tunnelling FET, (TFET); a gate controlled Esaki Diode: + Gate Gate p + n + p + n + n - n -

The Gate Controlled Esaki Diode: The sub-threshold slope for tunneling depends on the steepness of the band-edges: E F Current band to band tunneling Sharp Step Diffusion current Bias Voltage

The optical absorption coefficient, α(hν), of Si at 300K, in the vicinity of the band edge. The Urbach edge grows as: α(hν) ~ exp{(hν-e g )/E o }, where the E o parameter is a type of sub-threshold slope. E o ~ 10meV for Silicon Tom Tiedje, Eli Yablonovitch, George D. Cody, and Bonnie G. Brooks IEEE Trans. On Elec. Dev., VOL. ED-31, NO. 5, p. 711 (MAY 1984) It's good, but it should be better. We need to search for materials with steeper band-edges!

Why aren't the band edges sharp? thermal strains & Deformation Potentials E E c E v Energy level fluctuations = (% strain) Deformation Potential Energy level fluctuations = 0.5% 25meV 5eV

Surprisingly, Si & Ge have opposite sign Deformation Potentials: Conduction Band Energy (mev) 900 800 700 600 500 0.0 0.2 0.4 0.6 0.8 1.0 Si L X Si 1-x Ge x Short- Period L&X Superlattice Ge Deformation Potentials: Si : Ge : Ξ Δ d Ξ L d 1 + Ξ 3 1 + Ξ 3 Δ u L u = 4.18 ev = 1.54 ev opposite sign Cancelling the effects of the thermal sound waves might be possible: Deformation Potential Engineering?

Construct the channel as a short-period Si-Ge superlattice: Gate p + n + n - Si Ge The challenge is to cancel the effects of: isotropic strain uni-axial strain shear strain

If V dd is lower, then the current drive requirements go down too! Some of the tunneling transistors have a relatively low current drive capability: Current Density Requirement: For conventional transistors: 1 milli-amp / micron If a lower operating voltage is achieved, then the current required to charge the wires would also be lower. A Figure-of-Merit that allows for this possibility would be expressed as: 1 milli-siemen / micron A current density as low as 1 micro-amp / micron would be acceptable, if the operating voltage were as low as 1 milli-volt.

Recommendations: 1. Milli-Volt powering should be regarded as a Goal for future electronic switching devices. 2. There would be both an immediate benefit, as well as a benefit at the end of the roadmap. 3. Band edge steepness is poorly known, and should be investigated for a number of semiconductors and semi-metals. 4. The full range of technology options should be included.