As part of my ongoing deep dives into different portions of the climate solution space, I’ve been working my way through grid energy storage solutions. That led to my recent pair of broad articles on grid storage, Grid Storage Winners Part 1: Assessing The Major Technologies, and Grid Storage Winners Part 2: How Much Of Which Storage By When? The multi-factorial assessment found that lithium-ion batteries will have a smaller role than many assume, that closed-loop pumped hydro storage (a subject I’ve published on many times) would be a very large part of the solution, and that redox flow batteries would be second only to pumped hydro in global application.

As a teaser, Agora Energy Technologies’ CO2-based redox flow battery technology has won the 2021 Hello Tomorrow global deeptech competition across 5,000 entrants from 128 countries, the just announced 2021 Keeling Curve Prize and the CEO has been made a 2021 Cartier Women’s Initiative Fellow in science and technology. More on Agora in subsequent articles.

This ARPA-E sourced diagram of a Harvard flow battery is sufficient to start the discussion. The “flow” in flow batteries is for the movement of liquids through two chambers separated by a polymer membrane that allows the passage of subatomic particles between the chemicals.

There are often four tanks, not two, so this diagram represents one of the models. The pumps push liquids from a tank or two through the chambers into the other tank or two. As they pass through the chambers over the membrane in the presence of electricity, charged particles move from one chamber to the other through the membrane and the liquids’ chemical composition changes. What ends up in the storage tank(s) are different chemicals than what was being pushed through, and the new chemicals have stored a charge.

Reversing the pumps pushes the new chemicals back through the chambers and the charged particles migrate back through the membrane, reversing the chemical process and releasing electricity. In closed-loop flow batteries, you end up with exactly the same chemicals you started with and can repeat the process as many times as you like. Open-loop flow batteries raise very interesting possibilities, and more on that later.

For a sense of scale, a MWh of storage requires typically tons of liquid, but MWh storage in lithium-ion batteries weigh a lot too. An 85 KWh Tesla Model S battery weighs 540 kg or 1,200 lbs, so a MWh version would weigh around six tons. Also contextually, pumped hydro sees examples such as a gigaliter of water for a GWh of storage, suggesting tens of thousands of tons of water for a MWh. Energy storage requires mass.

A flow battery will look like a shipping container or small building surrounded by two to four large tanks, pumping equipment, and electrical grid connection and electricity-management components. It will look more like a chemical plant, not a battery, and there’s a thread there that I will pull on in a subsequent article.

The chambers and membranes have limitations in terms of scale. This isn’t one chamber and one membrane, but many chambers and membranes. It’s not one set of tubes leading to a single chamber, it’s a lot of sets of tubes leading to a lot of chambers. 

Once again, the analogy can be made to Tesla’s battery packs. They use a lot of small battery cells connected in series to achieve the voltage required and then in parallel to create the capacity required. The difference is that the contents of the lithium-ion battery remain in place until they degrade and battery capacity is lost and the individual cells have to be replaced entirely, but the flow batteries use chemicals which are easy to replenish as necessary.

Scale is also important. While Tesla’s individual battery cells have increased in size, they are still the size of a finger. You can hold a lot of them in one hand. They look like bigger versions of the AA batteries we put into our handheld electrical devices like razors and flashlights. There are about 4,400 of them in a single Tesla Model 3 to achieve the kWh of storage that a car requires to drive hundreds of kilometers.

Flow battery cells, on the other hand, are much bigger as individual components. Agora’s founders tell me that their individual cells will be scaled to 0.5 meters by 0.5 meters square and 1.0  centimeters thick on the inside, and a bit more than that with structural components and hose and electrical fittings. That’s 1.6 ft by 1.6 ft and perhaps half an inch internally, and perhaps 2-4 inches thick with the remainder. Their cells will have a capacity of processing various flow rates of liquid electrolyte, and room for the gaseous CO2 as well, a unique aspect of their technology.

The scale is part of what makes flow batteries interesting. Let’s take a brief digression into vertical vs horizontal scaling, or scaling up vs scaling by numbers, as my chemical plant engineer collaborator Paul Martin says they refer to it in that industry.

Vertical scaling makes individual components of a system bigger and more powerful. In computing, it leads to mainframes. In energy, it leads to GW nameplate capacity nuclear and coal plants. But vertical scaling above a certain point turns into a lot of engineering at the point of construction. Things get to a scale where they can’t be shipped, so they are delivered broken down into an often complicated 3D jigsaw puzzle of components that have to go together in a certain order with skilled resources assembling them. The history of failures of nuclear plants to be delivered on time and budget is testimony to the challenges of vertical scaling. In other words, scaling something up is good until it isn’t good any more.

Horizontal scaling, on the other hand, uses a lot of small, identical components to create the same output as a single vertically scaled component. In computing, that’s distributed server technology. In energy, that’s wind and solar farms. With horizontal scaling comes manufacturability of the individual identical components including factory quality control. And it provides for very standardized distribution and highly parallelized modular construction at the sites. 

This variance between the massively vertically scaled central power stations vs the massively horizontally scaled wind and solar farms is a very poorly understood competitive differentiator in the energy industry. It’s part of the reason, in my opinion, why energy analysts failed miserably to understand how much wind and solar were going to eat coal and nuclear plants. The analysts in that industry had no context for horizontal scaling and the enormous economies of that type of scaling. After all, there are only about 500 working nuclear plants in the world. A GW of wind energy capacity might have 400 wind turbines, and given capacity factors a wind farm that’s equivalent to a single nuclear plant might have 800 wind turbines. That’s an awful lot more identical manufactured components with identical templated assembly and a lot of room for optimization at every step of the supply and construction chain. 

Solar panels, of course, are even more horizontally scaled than wind turbines, with 200 kW panels a meter long weighing 30 kg or so, but tens of thousands of them. Solar panels are like chopsticks, manufacturable in massive volumes easily and cheaply, easy to stack in containers, easy to ship around the world, and easy for teams of humans to put on racks.

Horizontal scaling comes with its own challenges. If the components are too small, say the size of a human finger, it takes a lot of connections to achieve a very large output. That doesn’t make them ineffective or inefficient as the Tesla example shows very clearly, but it does suggest that there are advantages to scaling in both directions. The plummeting cost of solar shows that meso-scale objects that are manipulable by individual humans, that come in very regular dimensions and that have simple assembly processes on site have a lot of advantages.

Which is where we return to flow batteries. That 0.5m x 0.5m x 1 cm cell is about the size of a blade server in computing. Without liquid, it will weigh perhaps 0.5 to 0.8 kg approximately since most of it is carbon-based material, polymer membrane, and plastic. In other words, it’s in the scale of the devices that run the internet already and it’s in the scale where it’s easy for humans to manufacture, distribute and assemble in large volumes.

This image from a Sandia Labs presentation on energy storage starts to cement this. If I had said this was a server rack set and associated equipment, you would have believed me. But this is the type of thing you will see with flow batteries. The cells will slot into standard racks. They will have standard connections. Semi-skilled labor will be able to connect them together following Ikea-class instructions. There will be a lot fewer connections for a given scale of storage than using Tesla’s finger-sized batteries.

And the other aspect of flow batteries, that they have externalized chemical loads instead of the embodied chemistry of lithium-ion cells, has advantages as well. They are much lighter per unit when shipped, and completely inert. The chemicals will be stored in bog-standard industrial tanks, commoditized components that are easily shipped and assembled as well. Shipping tanks of chemicals is a very commoditized market as well.

This is not to say by any stretch that Tesla’s Powerpack is inferior, it just has different characteristics that mean its economics for manufacturing and distribution are different. The individual batteries are manufactured in highly automated factories and are very amenable to automated packing and distribution. Getting them all wired together into battery packs for cars or grid storage is also automatable. And they don’t need plumbing at point of installation, which is an advantage, just wiring. They also have a great deal more flexibility of use. I’ve seen proposals for flow batteries in cars, which is even less intelligent than putting hydrogen fuel cells in cars, although not as bad as Saudi Aramco’s tailpipe carbon capture nonsense.

One of my relevant recent experiences was working on a 6-year technology strategy for a global electronics manufacturer. (Neat trivia: they built a factory in Laos for the cheap land and labor, but strung their own internet cable across a bridge from Vietnam to gain stable connectivity of sufficient bandwidth.) Among other things, they build fully loaded racks of blade servers and components, built and tested in one of their Asian factories for major brands like HP and IBM. This suggests that flow batteries will be delivered to sites the same way fairly rapidly. They will be assembled into individual racks in factories, have strong quality control as manufactured objects, then be wrapped and shipped in standard shipping containers to sites where the racks will be mounted and the connections made. It’s likely that all of the plumbing and wiring for each rack will converge into one electrical input and output and two to four plumbing inputs and outputs. Depending on optimization, racks could contain dozens of pre-wired and pre-plumbed cells, ready to be placed in rows in a lightly air conditioned building.

And there’s a further thought down this optimization path. One of the things that Google does with its data centers is to turn them into container farms. They take standard shipping containers, outfit them with racks, stick blade servers in the racks, put in all of the air conditioning, power and connectivity components, then ship them using standard container transshipment distribution to leveled fields where there’s cheap electricity near dams. They drop an entire container, hook it up from the outside to power and the internet, and walk away. When the container’s servers degrade to a sufficient level, they pick up the entire container and ship it back to the factory for replacement and refurbishment. That’s likely the end point of flow battery optimization as well.

Flow batteries have useful operational characteristics as well. Because they don’t have the chemicals embodied in the cell and because the chemistry is sufficiently different, they don’t have recharging limitations. Tens of thousands of cycles are trivial for flow batteries, which is an advantage for grid storage. 

That doesn’t mean that they last forever or that they don’t have other constraints. Common chemistries such as vanadium redox flow batteries use toxic and acidic chemicals, so the transportation and siting comes with health and safety constraints. The corrosiveness needs to be engineered for. The membranes and catalysts degrade over time as well, and need to be replaced.

But maintenance can be easy as well. Because of the serial and parallel nature of a flow battery, you can shut down a string of racks and only reduce the overall capacity, not disable the storage in total. Because of the human scale of the cells, you can foresee a maintenance person pulling a cell out of a rack, putting a new cell in and shipping the cell off for refurbishment. And of course the end model of shipping containers going back to the factory is viable as well.

Flow battery characteristics typically mean that they are most cost effective at grid or major facility scales of storage. Per my assessment, most are useful for mid-duration storage of 6 to 48 hours. This means that they overlap nicely with the load shifting of more expensive Tesla-approach lithium-ion storage for in-day, fast response grid balancing and with 1-21 day storage in GWh-scale pumped hydro. I see each of the storage technologies have a place to play on well balanced grids.

This is part of a series of articles I’ll be publishing around this technology. At least two will be devoted to Agora Energy, as they have a unique chemistry and model based on a couple of fundamental insights which gives them what appears to me to be a very strong advantage in two different domains. 

Full disclosure. I have a professional relationship with Agora as a strategic advisor and Board observer. I did an initial strategy session with Agora about their redox flow battery technology in late 2019 and was blown away by what they had in hand, and my formal role with the firm started at the beginning of 2021. I commit to being as objective and honest as always, but be aware of my affiliation.