In the movie “Diamonds are Forever”,  

James Covalent Bond is chasing some  missing South African diamonds.

He discovers that his nemesis Blofeld  is cooking up an evil plot to use those  

diamonds for a big space laser  that he intended to auction off.

I find the plot impractical. Blofeld  should have instead concentrated his  

criminal organization’s resources  on developing diamond transistors!

Had he succeeded, he might have  reaped far greater profits than  

with some silly space laser. In this  video, we look at a very hard gate.

## Silicon Sucks

There are many good things about silicon  that make it the most used semiconductor,  

but it is far from perfect.

First is silicon's bandgap, or the  amount of energy needed to excite  

the element's electrons from a resting to  energized state. The wider the bandgap,  

the more energy needed to excite those electrons.

Silicon's bandgap is just 1.12 electron volts,  which is relatively narrow. The upside of such  

a narrow bandgap is that we do not need such  a high voltage to switch a silicon transistor.

But it also means that heat can cause that  transistor to inadvertently switch. With  

more heat, charge carriers can punch through  the gate and generate a small leakage current.

This current dissipates more heat,  

which worsens the leakage and makes even  more heat. A negative feedback loop.

This is why we need to be careful about how many  

silicon transistors within a given  area are turned on during operations.

Silicon also has a relatively low breakdown field.  If exposed to a strong enough electric field,  

the silicon might lose its insulating properties  - causing large currents to burst through.

Think of water going through a pipe. If the  water pressure gets too high, the pipe bursts.  

So silicon isn’t good to use in situations  of high heat and strong electric fields.

Like for example, the inverters in EVs that  convert DC to AC power for the EV's electric  

motors to use. These have to withstand high  voltages in the range of 400 volts or more.

Or RF amplifiers used to amplify high  frequency signals of anywhere from  

1-6 gigahertz for 5G base stations. Yet  higher frequencies for millimeterWave.

## Going Big Bandgap

So for these, people have explored what  are called wider bandgap materials.

Called as such because their bandgaps are  wider than silicon’s. There are a number  

of these known in the industry.  Let me go through a few of them.

Indium phosphide and gallium arsenide have  

somewhat higher bandgaps - 1.35 and  1.42 electron volts respectively.

The bandgaps of silicon carbide and  

gallium nitride are even wider  at 3.26 and 3.4 electron volts.

But we need to consider more than just the  raw bandgap. We must also consider economics,  

manufacturability and other things.

For instance, the bandgaps of Indium phosphide  and gallium arsenide are not that much higher  

than silicon's. Moreover, they are both  toxic to humans and on the rarer side.

Silicon carbide has relatively low carrier  mobility. Meaning that charge carriers  

like electron and electron holes do  not travel very fast through them.

This is partly because silicon carbide’s oxide is  

silicon dioxide. The interactions  between the silicon, oxygen,  

and carbon atoms create defects called traps  that can impede the charge carriers' movement.

Gallium nitride is nice and widely used in  the LED and optical industries. However it  

gets unstable at high switching frequencies  - leading to unpredictable behavior. And when  

exposed to the high heat and powerful electric  fields, it falls apart faster than silicon.

So while industries worth billions  of dollars have emerged around these  

high-bandgap materials, research continues for  other materials with possibly more to offer.

In the late 1980s and early 1990s,  people have explored what we call  

ultra-wide bandgap materials like aluminium  nitride, gallium oxide, and of course diamond.

## Diamond Rocks

Diamond rocks. And yes, it  is also a semiconductor.

Diamond's bandgap is absolutely massive compared  to silicon's: 5.5 electron volts compared to 1.12.

And its breakdown field of  10 megavolts per centimeter  

can be up to 33 times higher than silicon's.

And its charge carrier mobility  is good, especially very pure,  

"electronic-grade" diamond. Mobilities have  been measured at 3 times that of silicon  

for electrons and up to 9 times for electron  holes. Though there are nuances to consider.

Perhaps most useful of all is diamond's  unparalleled thermal conductivity. Thanks to  

strong covalent bonds which allow heat-carrying  phonons to propagate without interference,  

Diamond's thermal conductivity is  about 2,200 watts per meter-Kelvin.

That is amongst the highest of all the known  semiconductors. It blows away silicon's 150  

watts per meter-Kelvin and exceeds copper's 400  watts per meter-Kelvin and silicon carbide's 370.

So the advantages of this in a field effect  transistor are pretty dead obvious. People  

are already experimenting with high electron  mobility transistors made with diamond heat  

sinks beneath them. A diamond transistor has  that unparalleled heat sink built right in.

This means that we can dissipate far  more power through diamond at the same  

temperature than with silicon and other materials.  

Enabling very high densities of active  diamond transistors or less cooling.

There are other things too. Diamond is also quite  resistant to radiation, because of the tightness  

of the bonds. Such rad-hard qualities makes it  suitable for extreme conditions like space travel.

Perhaps the best thing is how diamond combines all  of these properties - high thermal conductivity,  

high breakdown voltage, radiation-hardness, and  high carrier mobility - in a single package.

No other material is like this,  and that makes it perfect for  

certain high-performance niches within the  $55-60 billion power electronics market.

Okay that is all the good stuff. The  reasons to care. Now here comes the catch.  

We use diamonds in industry, taking advantage  of its hardness or optical properties. But we  

do not have diamond transistors because  those are very hard to make. Ha ha.

Humans first fabricated a diamond-based  semiconducting device in the late 1950s,  

when M. Bell used naturally P-typed diamond to  

create power diodes. But progress  then stalled for several decades.

## Synthesis

The first problem with diamond  is making a lot of it, cheaply.

We can produce 12-inch wide silicon  wafers at scale by pulling crystals  

out of a molten silicon melt and cutting them.

The Czochralski method, yes. Moreover, those  wafers are single-crystal - meaning that we  

have a continuous, unbroken  crystal lattice of silicon.

Crystal defects - defined as dislocations of the  atoms, voids in the structure, or disruptions  

in the planes - can interfere with the charge  carriers and impede the device's performance.

Czochralski is well known and works.  But it cannot be applied to diamond  

because we cannot produce a vat of diamond "melt".

Unless you put it under untenable  high pressure conditions, molten  

diamond turns into graphite, which we cannot pull.

So back in the 1950s, the only way to  synthetically produce diamond was by replicating  

the high pressure and high temperature conditions  of the earth's insides, called the HPHT method.

HPHT involves putting carbon under a  massive piston made from super-hard  

alloy - subjecting it to conditions  of 5.5 gigapascals - about as much  

pressure as my dad exerts during SATs  - at 1,300 to 1,400 degrees Celsius.

HPHT-produced diamond crystals have few  lattice defects, which is good. But there  

are three problems. The first being that  the diamonds are too small to do much with.

And two, HPHT introduces unwanted impurities into  the diamond from its surroundings - like boron,  

cobalt, and nitrogen. Ergo why  so many HPHT diamonds are yellow,  

they have within them 10-100 parts  per million nitrogen impurities.

The third issue is that the HPHT method itself  is expensive and high-maintenance. Thus,  

a new method has emerged: A form of chemical  vapor deposition called Microwave Plasma CVD.

This involves injecting a mix  of precursor gases - methane  

about 97% diluted in hydrogen - into a chamber.

Powerful microwaves then strip  the gas atoms of their electrons,  

forming a plasma. The methane  decomposes into carbon and hydrogen.

The free carbon atoms then form chemical  reactions on the surface of a seed diamond,  

building a diamond layer vertically on top of it.

The fact that we use a seed diamond makes  this MPCVD process a homoepitaxy - epitaxy  

on the same. We are essentially extending the  seed diamond's existing crystal structure.

Commercialized in the 1980s, MPCVD is the  dominant method of producing gem-quality  

diamond at scale. I once visited a MPCVD  laboratory in Taichung - trying to produce  

electronic-grade synthetic diamond - and  thought it was pretty cool technology.

## No Big Diamond Wafers

However, more work needs to be done.  The first problem is getting the MPCVD  

growth rate up without compromising  on the quality of the grown crystal.

The microwave plasmas are inherently non-uniform,  

and their flow over the growing  diamond surface changes due to  

subtle things like even the presence of  gaps or edges on the substrate-carrier.

So areas of the wafer experience  different conditions - leading to  

uneven quality. And carbon nucleates  well on things, which makes diamond  

particularly apt to form as many small  crystals separated by grain boundaries.

We call this polycrystalline diamond,  

and it is littered with grain boundaries  and other boundary defects that impede the  

travel of charge carriers and negate the  material's inherent electronic advantages.

A critical issue is the lack of large  enough single-crystal seed diamonds.  

Most seeds are produced with HPHT,  

which grants us the single-crystal structure  we want but inherently limits their size.

And the complex dynamics of the gas  precursors inside the chamber mean  

that single-crystal diamond growth via MPCVD is  slow: 75 micrometers per hour, without nitrogen.

About twice that with nitrogen, though  nitrogen doping only goes so far because  

too much can lead to deficiencies like  what are called nitrogen vacancies.

I recall from my visit to  the MPCVD lab them saying  

that a 2-carat gem-quality diamond  might take about 2 weeks to produce.

The target width to go commercial is about  4-inches with a killer defect rate of 1 per  

10 square centimeters. But single-crystal  wafers just 10 millimeters wide can cost  

up to 10,000 times of equivalent  silicon. We are far from ready.

And by the way. Interestingly enough, pure,  electronic-grade diamond wafers are black colored.

## Trying to Get Big

Improving the growth rates  of the MPCVD method is tough.

That is because it is a complex reaction, and  isolating the impact of one factor requires a  

lot of runs. Though it can be improved by  running multiple tests at the same time.

Getting to 4 inches - or even just two -  requires that we expand beyond the seeds'  

physical dimensions. Tried methods including  using CVD to grow diamond laterally around  

the seed like a mold. This increases our  size by 1-2 times in a single run. Good,  

though sadly still far short of our goal.

Another idea has been to grow single-crystal  diamond on the sides of the seeds. This is  

interesting, but generally does not work well  because the seeds' sides are not that big.

The method that I saw but did not recognize  back in Taichung - is the mosaic method. We  

take multiple seeds and put them together  in a single "mosaic", like a window pane.

The diamond grows over the mosaic, fusing  together to create a larger piece that we  

can lift off the mosaic. The quality is quite good  since we are using single-crystal diamond seeds.

A side-channel has been trying to make  non-diamond seeds work - something  

called heteroepitaxy. If pure diamond does not  work, then let's try something else entirely.

The best candidate seems to be - surprisingly  enough - Iridium. And not just a slab of it,  

very special iridium layers deposited on  top of metal oxide and silicon substrates  

for cost and thermal reasons. Consensus  is that this is best way to eventually get  

to 4-inch single-crystal wafers,  but issues like defects persist.

## Doping

Another major issue is doping.  This is an essential part of the  

transistor production process. Without  it, the material is electrically inert.

In the case of silicon, we can  produce electrically active,  

N or P-type silicon by injecting  impurity atoms like phosphorus or  

boron into the material. Nowadays this is  done with a method called ion implantation.

We want to be able to do the same for diamond,  

and have it work just like a  regular transistor, but it does not.

If we want to imagine the transistor as  transporting water from the source to the  

drain through the channel, we need to make  sure that there is water in that source.

With silicon, there is plenty of water  - meaning charge carriers - available.  

The boron and phosphorous atoms  inside the silicon are willing to  

give up their charge carriers when given  just a small amount of thermal energy.

At about 0.045 electron volts for both boron and  phosphorous, silicon's ionization energy threshold  

is low enough - or shallow enough, if we are to  use industry lingo - that room temperature works.

But the threshold for doped diamond is multiples  higher. Err. The correct lingo is to say,  

"deeper". Boron's is 0.36 electron volts  and phosphorous's is even deeper at 0.57.

Thus at room temperature, only a small  percentage of the boron atoms give up  

their charge carriers. And so unless the diamond  transistor is subjected to high temperatures,  

you get too few charge carriers for  the transistor to run efficiently.

That's great if we want to run our diamond  transistor computer on the surface of  

Venus - and it legit might perform really  well there - but we are talking about Earth.

A few alternative dopants exist but the situation  there is even worse. Practically speaking,  

it has got to be boron and phosphorous.  We have to figure out a way around.

## Struggling with Boron

Let us start with boron, which  is used to create P-type diamond.

First, they tried to overcome this low efficiency  with brute force. If there is not enough boron  

atoms giving up their electron holes? Then  let us just add more into the diamond lattice!

Unfortunately, too much doping causes the  diamond to act less like a semiconductor  

and more like a metal. They use the phrase  "degenerate semiconductor", which is evocative.

So they then tried something called delta-doping,  which is where you create a very thin layer of  

heavily doped boron diamond sandwiched  between two layers of normal diamond.

But scientists have struggled  to consistently do this,  

with methods like ion implantation known to  cause serious damage to the diamond lattice.

Note how efforts centered on boron and  P-type diamond. In order to create a  

silicon CMOS-style arrangement, then we  need the N-type diamond transistor too.

Unfortunately the challenges  there are even more daunting,  

since the dopant phosphorous activation energy  is a magnitude deeper than boron. And the only  

other option is nitrogen - whose activation  energy is a freakin' 1.7 electron volts.

## Hydrogen-Termination

The doping issues around diamond were so bad that  it derailed technological progress in the field,  

with investment and effort going  to silicon carbide and others.

But in 1989, a chance experiment with CVD  diamond discovered that if we terminated  

the dangling carbon bonds on the diamond surface,  

we can produce a conductive surface a  trillion times higher than expected.

What does this terminology mean?  The carbon atoms on the diamond  

surface has its bonds left dangling out  there. These dangling bonds are unstable  

and can grab passing charge carriers like  electrons and interfere with their travel.

Kind of like how anemones grab things  in the ocean to eat them. To terminate  

these dangling bonds means to attach  an atom to them to "satisfy" them.

The result of this termination is a very thin  layer of diamond that electron holes can travel  

through - giving us P-type conductive diamond  while sidestepping the whole boron doping problem.

It triggered a lot of research in the 1990s  to confirm. Upon confirmation that it was  

indeed the hydrogen termination doing this, we  were able to finally make switching devices.

## The MESFETs

The first serious field effect  transistor with diamond emerged in 1994.

They demonstrated a MESFET - a metal  semiconductor field effect transistor.

Meaning a type of layered transistor with the  source, drain, and gate all made from a metal.

We start with the substrate, made from  un-doped HPHT single-crystal diamond.

A layer of diamond was grown on top of that with  MPCVD. Hydrogen-termination via microwave-enabled  

hydrogen plasma is then used to make that  layer P-type conductive, moving electron holes.

Then two pieces of gold is deposited on top  to serve as the source and drain. In between,  

we have an aluminium gate. You  end up with a working MESFET,  

albeit rather quirkily and  with slower charge carriers.

Over the years, more MESFETs and variants  have been made with modifications added for  

improvement. Like, hydrogen-termination  being replaced with oxygen-termination,  

which is more stable in air.

But major challenges remain. For instance,  the transistors in modern power electronics  

are vertical - with the source and drain  on opposite sides of the wafer - because it  

lets a power current flow through the bulk of the  device for heat, efficiency, and voltage reasons.

In the diamond MESFET, power flows only  through the thin oxygen or hydrogen terminated  

layer at the surface of the device. Changing this  means fixing the doping problem again. So yes,  

we are rather far from diamond  transistors that can compete  

with silicon carbide or GaN power electronics.

I should note that we do not have to achieve  a full diamond transistor to get its benefits.  

I briefly mentioned earlier the  Gate-after-diamond method, where a  

layer of diamond is laid down to serve as a  heat sink for certain transistors that run hot,  

like High Electron Mobility Transistors,  or HEMTs. Maybe that is the future.

## Conclusion

The idea of a diamond transistor is  fun. And there is real reason to say  

that diamond can produce the ultimate  hard-core power electronics device.

But as it so often turns out in semiconductors,  

superior material properties is  just a small part of the equation.  

It needs to be manufacturable, scalable and  economic. Hopefully with existing hardware.

Even after overcoming diamond's  existing size and doping shortcomings,  

it might take years more to bring something  to market. Like with silicon carbide,  

where 4-inch wafers were ready by 1999 but  no devices came out until CREE's in 2011.

Nevertheless, the market’s growing needs and the  

material’s own inherent allure will continue  to drive more research in the years to come.