Energy harvesting – a real-world case

Jimmy_Xu@brown.edu; 13 Fevrier 2024 Mardi 13h30-17h45

• AI & cloud-computing in a race to ‘baseload power’ at scale & in time

• A potential solution, with EU in lead

• ‘Devils in the details’

• Think big, act small (at nanoscale, at molecular scale, at MW scale)

Your task (homework, due in one week & by email to
Jimmy_Xu@brown.edu on Feb 20, 2024)
Based on the lecture, the lecture notes, and the relevant literatures you can find,
write an report on:
1. Your understanding of the need for accessing power in the cloud- and AI-computing, its
scale, its urgency, its driver, its potential solution(s).
2. Your assessment of the leading potential solution(s) mentioned or your own
alternative/better idea(s)/solution(s)
3. Estimate the required p- and n-doping doping levels in silicon that would be required for
the water-splitting approach reported in this paper:
https://doi.org/10.1038/s41586-022-05549-5; Water splitting with silicon p–i–n
superlattices suspended in solution, T.S. Teitsworth, et al, 270 | Nature | Vol 614 | 9
February 2023
4. Assess the likelihood of its having been achieved in this paper based on the data they
presented and your analysis of the pn-junction built-in potential and its corresponding I-V
current-voltage dependence.
5. Propose and describe an alternative design or modification to improve the chance of
achieving their desired functionality (i.e. stacking up the Voc via the Esaki tunneling).
Starting with of a ‘case study’ on the investment side:

Data center investment note

Jimmy Xu
(22-Janv, 2024)
Investors’ perspectives – debt lenders (
Capital, Nord/LB…)
Sumitomo Mitsui Banking Corporation, Generate

• AI is the driver for demand that we have never seen before.

• We are seeing huge mismatch between supply and demand.
• It “has clearly brought a level of demand, ambition, and opportunity
above and beyond what was being discussed and contemplated a
year ago,” Schwartz, CyrusOne (hyperscaler-focused)

• Speed to power is the driver. We seeing long lead time for access to
capacity
• Power availability is a big factor – everybody in the industry is talking
about. On-site power, microgrids…
• Energy issues create investment hot-spots
• More investments in Frankford, Lisbon, Milan, Madrid, Warsaw,…for AI and low latency

• We are seeing space and power optionality constraint.

• “No space with power” in F.L.A.P.D. (developers are looking for
temporary gas-fired power to bridge that gap). More scrutiny from regulators are
adding barriers via demanding PUE and renewables.

• This may drive ‘data gravity’ (hyperscalers) to relocate capacities to

new areas where there are easier access to more sustainable power,
by moving mission and latency critical to key hubs (and having AI
‘anywhere’, in ’less cool hubs’)
• Looking from the energy perspective: the #1 issue for data
centers is “access to power” – MW capacity and within a
reasonable timeframe.
• We are seeing interconnection lead times extremely long. The
longest we heard is 2037 to get access to power.
• Speed to power is the driver, that is the key thing to solve for.
Power is less certain than ever. On-site power is one solution

Broadly, we are all for sustainable digital infrastructure grid of the

future. The future is in distributed infrastructure.
Options for data centers to secure power
1) hyperscalers are looking into building new locations, perhaps have a 5-10 years lead time
to interconnections. They are looking at bridge solutions, like fuel cells (Bloom), by deploying
on-site power plant to bridge the time until interconnection gets there, or even supplement
that power once finally interconnected. Demand for fuel cells has gone quite high, almost
exponentially. For context, 3 years ago, the biggest deal was 20MW. Now, between 100MW
and 250MW for fuel cells. (They are looking for maximum optionality. They may have 10-15
sites in the pipeline over the next 10 years. They’d like to have predictable time to power.
How to rapidly deploy on-site power plants and meet the demand? We can underwrite to
the short term, the PPA term, say 6-year term to bridge that time to interconnection. We can
then move the fuel cell plant to another site to get the rotational element, to help them get
the speed to market faster.
2) is to collocate next to the power plant, construct the substation right next to the power
plant and bypassing the grid infrastructure, and do deals with independent power producers.
It started with the bitcoin miner, is lending itself nicely to a lot AI players (deploying
modularly to get access to MW, buying distressed bitcoin miners…)
3)....Offer me ”Powered Land package”, developers!
 #1 factor is ”Speed to Market”.
During this transition, we have to become creative for energy infrastructure in
how we serve this demand, that has become proven as financeable.

Base principle – any DC needs baseload power.

Realistic option – either gas or nuclear.
Solar/Wind is just not there. You can make a case for geothermal, but there is
a long ramp. Hydro is location dependent, only if you are lucky...

Gas-feed fuel cells can bridge the time to interconnection, it has a path to
greener fuel, and can be deployed fast and then movable to new sites.

SMR – Scifi? Or reality? The biggest challenge is around the regulatory

environment. We are seeing a few companies that have large pipeline in parts
of Europe where the red-tape has been cut sufficiently that they can go to
deploy. In the USA, we don’t see it happening in the near term (i.e. 5 to 10
years).
Note: this is from someone specializing in the power sector, who also says “if not underwritable, it is not deployable”
Bloom fuel cell – a hydrogen power at scale
• Problem: data center needs ‘baseload’, ESG.
• Problem:
• transmission constraints,
• severe weather-driven resilience challenges,
• intermittent renewables,
• underperforming hydro assets,
• planned nuclear retirement…
• Example: AWS learned that the local utility could
not add capacity in time, and selected the onsite,
solid oxide fuel cell to provide distributed energy
to three proposed data center sites in Oregon
requiring 24.3 MW at each location.
• The fuel cells that convert natural gas to
electricity, with a 99% decrease in smog-forming
pollutants and water use and 50% lower carbon
emissions than the displaced grid power in the
region.
• The fuel cells have a minimal footprint or
infrastructure
Solid Oxide Fuel Cell. - BloomEnergy
• solid oxide fuel cells (SOFCs) hold the greatest potential of any fuel cell technology,
• due to their extremely high electrical efficiencies and low operating costs.
• It has been on every NASA space mission since 1960’s.
• SOFCs are likely to emerge as the fastest growing fuel cell segment over the next six
years (Bloom Energy 2019 )
• The steam that is produced in the reaction is recycled to reform the fuel. It does not require
water during normal operation.
• Thermal power plants require significant amounts of water for cooling, it is the number one use
of water in the U.S.
• To produce one megawatt per hour for a year, thermoelectric power generation for the U.S. grid
withdraws ~156 million gallons of water.
• The electrochemical process also generates the heat required to keep the fuel cell warm and
drive the reforming reaction process. As long as fuel and air are available, the fuel cells continue
converting chemical energy into electrical energy, providing an electric current directly at the
fuel cell site.
Operating at high temperatures inside the Energy Server (‘Bloom Box’)
ambient air enters the cathode side of the fuel cell.
Meanwhile, steam mixes with fuel (natural gas or biogas) entering from the anode side to produce reformed
fuel (via electrochemical reaction, not combustion, that is 50% less carbon emission).
As the reformed fuel crosses the anode, it attracts oxygen ions from the cathode.
The oxygen ions combine with the reformed fuel to produce electricity, steam, and carbon dioxide.
That is about one solution mostly in the USA

How about EU? … also on the move

would they be big enough, soon enough, good enough?

Example – wind/solar to hydrogen

EU countries led by Denmark plan to build “Energy
Islands” offshore with giant offshore wind farms Jan 10, 2024
• connect giant offshore wind farms (3-4GW first, 10GW later) to shore while
housing on-island hydrogen projects and data centers
• Addressing the problems of local objections to close-to-shore installations and
generally less powerful and more intermittent wind, while offshore the sea is
deeper and presents significant structural problems.
• The wind energy could also be converted to hydrogen via electrolyzers; these
could either be sited on the island itself, with the hydrogen piped back to shore,
or on the mainland, where the excess heat from the process could be fed into
district heating networks to heat homes.
• Danish Energy Agency (ENS) is leading a project to build two islands, one in the North Sea
and one in the Baltic Sea. The successful bidders will be announced next year (2024).
• "Denmark has a lot of coastline, and the North Sea particularly is a fantastic area for
offshore wind projects - very windy and with relatively shallow waters,"
Belgium
• A Belgian consortium, TM Edison, has won the tender from electricity transmission
system operator Elia for the construction of an artificial energy island, again in the
North Sea, but this time off the coast of Belgium.
• Princess Elisabeth Island will be located 45km from the coast within the Princess
Elisabeth wind zone. The idea is to connect a 3.5GW wind zone with the UK and
Denmark through the Nautilus and TritonLink interconnector projects, as well as
Belgium itself.
• The plan is to link all wind farms in the zone to the mainland by 2030, and ultimately
to generate 300GW of offshore electricity by 2050.
• Construction of the foundations is set to start next year and is expected to take two
and a half years, with the high-voltage infrastructure installed after that. It will, says
the team, be the world’s first artificial energy island to combine both direct current
(HVDC) and alternating current (HVAC) facilities.
EU solution may also address these problems

• Scale (10s, 100s MW)

• Societal: “I am all for solar and wind power, but not in my backyard”

• Intermittent solar/wind
• Fluctuating demand
Storing electron energy in chemical bonds
• The ‘grey color’ of hydrogen, and
• Fishery
• Costly electric power line to shore (wait, sure?)
The colors of hydrogen
‘C-suit question’: alternatives &
comparison ?

Credit: Linfeng&Fatemeh report for our first ‘Energy Harvesting’ class, 9 Janv, 2024; [data from
Parthasarathy and Narayanan (2014)]. Comment: for the C-suit it would need to be ~2024

As of 2022, commercial electrolysis requires around 53 kWh of electricity to produce one kg of hydrogen, which
holds 39.4 kWh (HHV) of energy @74% https://newatlas.com/energy/hysata-efficient-hydrogen-electrolysis/
So, hydrogen for ‘the rich’ ?
• Can we make hydrogen for ‘the poor’ too?

• This is the business of atoms, electrons, and energy

• It is in our space and

Isn’t it our business ?

Photoelectrical
water splitting
Jimmy Xu
Jan 11, 2024

Objective: Solar-chemical energy conversion/storage

Electrochemical water splitting
• Water electrolysis requires a minimum
potential difference of 1.23 volts, although
at that voltage external heat is also
required.
• Typically 1.5 volts is required.
• Electrolysis is rare in industrial applications
since hydrogen can be produced less
expensively from fossil fuels.
• As of 2022, commercial electrolysis
requires around 53 kWh of electricity to
produce one kg of hydrogen, which holds
39.4 kWh (HHV) of energy @74%
https://newatlas.com/energy/hysata-efficient-hydrogen-
electrolysis/
2x H2O  2xH2 + O2 Or,

+ = > mostly for ”O- reduction” (i.e. losing e- to anode)

1.23eV
Its dependence on
1.23
Energy = 1.23eV + ‘overpotential’

• Sources of the 1.23eV: electricity, light, heat, chemical reaction

• Sources of the overpotential: incomplete access to surface (e.g. bubbles,

density gradient, charge mutual repulsion, double-layer screening, partial
landing/adsorption…)

• Loss to recombination (needs to separate the split ions by separation

membrane, and other innovative plays of capillary forces for negative and
positive ions)
Degrees of freedom in/for engineering
• Local field
• Potential (accumulative to beyond 1.23eV)
• Energy injection (solar and/or wind, i.e. light, electricity, heat)
• And energy capturing and conversion to local fields
1. Local field, conversion and accumulation to exceed the energy
threshold
2. Material and structure engineering to lower the energy threshold
3. Material and structure engineering to ease the landing
(adsorption), removal of reaction products, separation , resupply
of H2O without hindering the separation/removal of H2 and O2
 Platinum  H vs O
F = - 5.65eV

https://www.universiteitleiden.nl/en/news/2019/05/we-finally-understand-ho
w-oxygen-reacts-on-platinum
https://www.pnas.org/doi/10.1073/pnas.1902846116
A process reversal – forming water w/o energy
• The H & O molecules diffuse towards the
surface and are adsorbed on the surface of
the pieces of platinum.
• The O& H molecules are held on the surface
of the platinum by temporary bonds formed
between these molecules and the platinum
Platinum is a transition metal & it has
vacant d orbitals that help the molecules to
form temporary bonds with it.
• The higher concentration of the molecules
on the surface of the catalyst makes it more
likely that the molecules will collide with
each other.
• This will result in bonds being broken & new
ones formed, i.e. a chemical reaction will
occur to form water molecules on the
surface of the platinum.
• Reaction is helped by the fact that the
covalent bonds holding the atoms together
are weakened by the formation of new
bonds between the reactants and the
surface of the catalyst.
Changing PH (chemists’ way of changing local charge–field)
A calling – for semiconductor engineers?
What can we do to manipulate the local field and charges?
At the molecular scale? Nanoscale?

We do this all the time! And, successfully! Better than anyone else!

Look at what we did with just one material – silicon!

• We made billions and billions in transistors, and
• In dollars, and now
• Our civilization depends on it!
Silicon
Also great for solar energy harvesting. Why so?
1. Vast infrastructure (material, fabrication, scalability, know-how)
2. Mature technologies (doping, contact, oxide, etching, …)
3. Bandgap 1.1 eV nearly-ideal for solar spectrum (in terms VocIoc, therefore fill-factor
FF at 300K vs solar spectrum)

• True of #1 and #2 for water splitting, as a form of energy conversion (from

radiation energy to chemical energy) and storage, but

• What about #3 ?
• comes just a little too short of > 1.23eV.
Problem with silicon in water splitting

• Threshold voltage  1.23eV + overpotential

• Platinum catalyst can lower the overpotential effectively but not to zero
• Silicon bandgap Eg = 1.1 eV
• Silicon PN junction built-in potential eV0 ~ < Eg

• This is true for 3D silicon bulk. Here are some research questions:
• With quantum confinement the effective Eg’ can be larger, and eV0 too?
• What is the built-in potential of a 1D or 2D silicon nanowire?
• Should the effective bandgap be higher due to quantization of both electrons
and holes?
• What would be the depletion width of a heavily doped 1D, 2D silicon pn
junction?
• we report Si-based PSRs by
synthesizing high-
photovoltage multijunction
Si nanowires (SiNWs)
• tunable photovoltages
exceeding 10 V were
observed under 1 sun
illumination.
• Spatioselective
photoelectrodeposition of
oxygen and hydrogen
evolution co-catalysts
enabled water splitting at
infrared wavelengths up to
approximately 1,050 nm,…
 A newly reported solution to the problem: stacking up pn-junctions

Eg Eg + DE

3D (diameter = 150-200nm >> exciton Bohr radius = 4.2nm) 1D (diameter < 4.2nm), Or, 2D
https://doi.org/10.1038/s41586-022-05549-5; Water splitting with silicon p–i–n superlattices suspended in solution, T.S. Teitsworth, et al, 270 | Nature | Vol 614 | 9 February 2023
Work function of elements (eV)
Ag 4.26 – 4.74 Al 4.06 – 4.26 As 3.75

Au 5.10 – 5.47 B ~4.45 Ba 2.52 – 2.70

Be 4.98 Bi 4.31 C ~5

Ca 2.87 Cd 4.08 Ce 2.9

Co 5 Cr 4.5 Cs 2.14

Cu 4.53 – 5.10 Eu 2.5 Fe: 4.67 – 4.81

Ga 4.32 Gd 2.90 Hf 3.90

Hg 4.475 In 4.09 Ir 5.00 – 5.67

K 2.29 La 3.5 Li 2.9

Lu ~3.3 Mg 3.66 Mn 4.1

Mo 4.36 – 4.95 Na 2.36 Nb 3.95 – 4.87

Nd 3.2 Ni 5.04 – 5.35 Os 5.93

Pb 4.25 Pd 5.22 – 5.60 Pt 5.12 – 5.93

Rb 2.261 Re 4.72 Rh 4.98

Ru 4.71 Sb 4.55 – 4.70 Sc 3.5

Se 5.9 Si 4.60 – 4.85 Sm 2.7

Sn 4.42 Sr ~2.59 Ta 4.00 – 4.80

Tb 3.00 Te 4.95 Th 3.4

Ti 4.33 Tl ~3.84 U 3.63 – 3.90

V 4.3 W 4.32 – 4.55 Y 3.1

Yb 2.60[15] Zn 3.63 – 4.9 Zr 4.05

https://en.wikipedia.org/wiki/Work_function
Ohmic and Schottky contact to p-type silicon
Ohmic:
• Heavily doped p+ silicon: Simple and low-cost but suffers from higher sheet resistance compared to metal contacts.
• Gold (Au): Often used due to its low work function and ease of processing. However, it requires additional steps for
stability.
• Aluminum (Al 4.2eV): Not ideal for ohmic contact on p-type due to its low work function forming a rectifying
barrier.
• Beryllium (Be) or Nickel (Ni 5.1eV) offer higher work functions for ohmic behavior.
• Silicides: Similar to n-type, TiSi2 and others can be used, but processing temperatures and specific dopants might
differ.

Schottky:
• Aluminum (Al 4.2eV): Forms a Schottky barrier with moderate height, suitable for general-purpose applications.
• Silver (Ag): Lower barrier height compared to Al, suitable for low-power devices with lower forward voltage drop.
Ohmic and Schottky contact to n-type silicon
Ohmic:
• Heavily doped n+ silicon: This is often used for simplicity and low cost, but suffers from higher sheet
resistance compared to metal contacts.
• Aluminum (Al ~4.2eV): Popular due to its low work function and ease of processing. However, it forms a
spiking interface with silicon, requiring additional steps for stability.
• Ti (4.33eV) and TiSi2: A silicide compound offering low resistance and good stability, but requires higher
processing temperatures.
• Other silicides: (Co >4.53eV) CoSi2 and MoSi2 offer alternative options with specific advantages depending
on application.

Schottky:
• Gold (Au 5.1eV): Forms a Schottky barrier with moderate height, often used for general-purpose
applications.
• Platinum (Pt 5.6eV ): Offers higher barrier height and lower leakage current, suitable for low-power
devices.
Issues and flawed hypothesis
• The turn-on voltage of a single pn-junction is low (<1V), which
suggests a low built-in potential, which in turn is consistent to likely
surface Fermi-level pinning.
• If the above is true, then it is hard to imagine a band-to-band
tunneling between n and p. The tunnel diode model explanation
might be misleading, or difficult to stand close scrutiny.

• Is it possible that the ‘shorting’ is via the surface (recombination)?

Structural engineering & physics
• Micrometer scale
• l & sub-l optics (resonant light absorption of targeted l-bands, near-field
intensification, ‘metamaterial’, optical-antenna, length, width, period, depth,
geometry, photon diffusion/trapping …e.g. local heating by ‘waste l-bands’)
• Fluidic transport (Laplace force, viscosity, surface, wetting, 1D, 2D, 3D
networks, directionality)
• Ionic transport and screening (Coulomb force, Debye length, double-layer,
ionically differential forces)
• 1D, 2D PN-junction 3D field-distribution, built-in potential, band alignment for
maximizing Voc, Isc…
• Sub-mm and mesoscopic scale
Structural engineering & physics
• Sub-mm and mesoscopic scale
• Hot-spot fields, dipolar, coupling to quadrupolar
• Spacing, interdigitated electrodes
• Easki-tunneling (3D)
• Quantum/resonant interband tunneling in 1D, 2D (Sweeny & Xu, R.Q. Yang & Xu etc.)
• 3D networks of Pt-silicon PN nano-particles with built-in directional fluidic transport and ion-separation
• Quantum scale
• Catalyst shape, edge, facet,
• H+ and O- landing, capturing, reduction
• Overpotential engineering
• Chemistry is beyond my know-how, colleagues?
• Scalability
• SOI
• Stamping or imprinting lithography
• Buried microfluidic channels and networks for thin-film portable H-fuel and O-fuel separation (resembles the
glucose test strip).  patent? And EU, France-US proposal?
Gas phase discomposing, and water splitting
Inspiration: Local field enhancement of efficiency, by 12x @ 8V/cm(field-ionization, de-
trapping to reduce loss of electrons to recombination in TiO2 defects) – Bruno
Grandidier’s work
Build-in local (e.g. surface) field via:
• pn-junction underlayer (e.g. lateral micro channels/stripes arrays)
• capacitive underlayer (e.g. supercapacitor, but with porous top contact surface for fringe field
penetration, e.g. capacitive square pad, or interdigitated electrodes on the same plane)
• ferroelectric underlayer (e.g. BaTiO3 and AlN); or
• sliding 2D vdW (inversion symmetry breaking, MoS2, InSe, GaSe, BN...)
Build-in self-powered (or Laplace-powered) reactants transport via:
• 1D, 2D (e.g. star, sunflower), 3D (‘tree’, ’bone’, porous matrix, concrete...) networks
• raise surface/volume ratio
• directional force (charge dependent, ion size dependent)
 Ferroelectricity via breaking inversion symmetry

10.1073/pnas.2115703118 M. Wu, J. Li, Sliding

ferroelectricity in 2D van der Waals materials: Related
physics and future opportunities,
Proc Natl Acad Sci U S A. 2021 Dec 14; 118(50):
e2115703118
Extending our reach
Extending energy-harvesting to ….
Beam-power, remote- and in-flight charging
The problem(s)
• There are many situations in which the direct use of electrical energy to power
electronic equipment is either not possible or not recommendable.
• Areas with these restrictions are generally called ‘‘exclusion regions’’ and exist in
applications needing to monitor and control processes under strict safety conditions
because of a high fire or explosion risk (refineries, mines, fuel tanks in space, and
aircrafts, etc.), the possible malfunction of sensitive electrical equipment in areas of
high electromagnetic noise (nuclear plants), the need of galvanic insulation (high
voltage lines and lighting-safe monitoring), etc.
• Alternative fields of application lie in the remote powering of rechargeable batteries
(medical applications), presence or flame detectors, rotating systems, and even for
powering satellites during periods of eclipse or Moon bases from Earth in the Moon
night.
• (Mass*Size)/Duration ratio, e.g. drone, electronic implants
A solution – Power Beam & Photovoltaic laser power converter
• The core of a PBL system is the photovoltaic laser power converter
(PVLPC), which transforms the laser light delivered through an optical
fiber into electricity.
• Recently, a PVLPC has demonstrated the highest efficiency for any
photovoltaic converter, i.e., 68.9% at a laser illumination of 858 nm.

(Joule 6, 340–368, February 16, 2022 ª 2021 Elsevier)

Organic alternative?
• Lightweight
• Scalable – vertically and laterally
• Conformal, fordable
• spray-able, coat-able, printable

• Your creative ideas/proposals?

 Organic solar cell
• Performance
• Why Au nanoparticles made Voc smaller and Isc larger
• In the random network of donor-acceptor junctions, is there a maximum Voc? Is it at the
percolative limit?
• Optics
• Grating – its periodicity can selectively enhance a narrow band. Its optimization has a few
accessible parameters (linewidth, spacing, and near-field zone)
• For harvesting solar energy over a broadband, consider fractals, and sunflower (Fibonacci spirals)
• Materials
• Extend to IR sensing-film (via ML?)
• Nonlinear optical properties (PM6 – asymmetry X(2), Y6-symmetry X(3))
• Thermoelectric (thermal power, Seebeck, lateral scalability, pn-junction thermoelectric, random
networks of pn-junctions and its maximum at the percolation limit? )
• Supercapacitor (the donor-acceptor junction is effectively a capacitor, could it be injected into and
form inside a porous structure to form a supercapacitor for energy storage)?
• Ink-jet printing, beam-power Photovoltaic converter