Atomic cobalt on nitrogen-doped graphene for hydrogen generation

If you want a clean hydrogen economy, you run headfirst into a stubborn bottleneck. We can split water and make hydrogen just fine in the lab, but at scale we lean on platinum to make the hydrogen evolution reaction hum. Platinum is great because it binds hydrogen just right and lets it go just as easily. It's also rare and expensive. So the question has been, can we get platinum-like behavior from something earth-abundant and do it without dumping in huge amounts of metal? One clever answer is to go small. Not nanoparticle small—single-atom small. Imagine pinning individual metal atoms on a solid support so that every atom is an active site. Nothing is buried. Nothing is wasted. You can also sculpt the immediate neighborhood around each atom—the coordination environment—to nudge its chemistry in your favor. That's the spirit behind a study from Fei and colleagues, who took cobalt, a far cheaper metal, and dispersed it as isolated atoms on nitrogen-doped graphene. They call it Co–NG, and the big idea is that nitrogen atoms in the graphene lattice act like tiny anchors, holding cobalt in place and tuning its activity for hydrogen evolution. The result isn't subtle. This material kicks off hydrogen production at an overpotential of about 30 millivolts in acid. For an earth-abundant catalyst on carbon, that's startlingly low. Overpotential, by the way, is the extra voltage you need to push the reaction beyond its thermodynamic limit. Less is better. And here, less is very much the point. What makes this so striking is that the catalyst uses only a trace of cobalt, held atom-by-atom in a carbon sheet, and yet it comes within reach of behaviors we usually associate with platinum. Let's talk about how they built it, because practicality matters if you're dreaming in gigawatts. The team started with graphene oxide—think of it as a chemically friendly, easily processed form of graphene—and mixed in a tiny amount of a cobalt salt. The ratio was extreme: roughly 135 parts graphene oxide to 1 part cobalt by weight. They then freeze-dried the slurry to keep the sheets from collapsing onto one another and heated the fluffy solid in ammonia to 750 degrees Celsius. That ammonia treatment is key. It dopes the graphene with nitrogen, and those nitrogen sites are where cobalt likes to sit. You can even make the catalyst as a free-standing paper by filtering the precursor and annealing, which hints at scalable, binder-free electrodes rather than just powder on a lab disk. What's actually in there? Chemically, the finished material is mostly carbon with a healthy dose of nitrogen and a whisper of cobalt: about 88 percent carbon, 8.5 percent nitrogen, and only 0.57 percent cobalt by atom. X-ray photoelectron spectroscopy sees cobalt in a cationic state and a nitrogen spectrum rich in pyridinic and metal-bound nitrogen—exactly the kinds of sites that would cradle a single cobalt atom. When Fei's team looked by high-angle annular dark field scanning transmission electron microscopy—basically, a way to light up heavy atoms on a light background—they saw bright dots two to three angstroms across scattered in the carbon matrix. Dots that size are individual atoms. Importantly, they were surrounded by lighter elements, the expected carbon and nitrogen. Now, it's one thing to see dots. It's another to prove those dots are cobalt atoms bonded to nitrogen and not lurking cobalt clusters you failed to catch. That's where extended X-ray absorption fine structure, or EXAFS, earns its keep. By examining how cobalt absorbs X-rays and how those signals interfere because of nearby atoms, you can deduce who cobalt's neighbors are and how far away they sit. In this case, the first strong shell shows up at a distance consistent with light atom neighbors, not cobalt-cobalt. And when they transform the EXAFS into a wavelet map—a way to separate contributions by distance and by scattering power—the signature shifts exactly where you'd expect for cobalt-nitrogen bonds. A control sample made without ammonia shows a different fingerprint that matches cobalt-carbon bonds. No cobalt-cobalt peak appears in the data for Co–NG. That's the tell. Atomically dispersed cobalt, coordinated to nitrogen, with no detectable metallic particles. Performance is where the model meets the road. In sulfuric acid, Co–NG starts catalyzing hydrogen around 30 millivolts of overpotential, defined at a tiny current density just to mark the onset. Push harder and the current shoots up. It needs about 70 millivolts to hit 1 milliamp per square centimeter and roughly 147 millivolts to reach 10 milliamps per square centimeter. Those are the kinds of current densities engineers pay attention to because they map onto real device operation. Almost all the current goes to hydrogen—gas chromatography puts the Faradaic efficiency near 100 percent—so the electrons aren't sneaking off to other side reactions. And compared with controls, the contrast is sharp: nitrogen-doped graphene without cobalt and cobalt on undoped graphene both need more than 200 millivolts just to wake up. The lesson is that the cobalt-nitrogen motif, created by that ammonia treatment, unlocks the active sites. How fast are the kinetics once the reaction is running? A Tafel plot, which relates how much extra voltage you need to how much current you get on a logarithmic scale, gives a quick read. Platinum on carbon shows a slope of about 31 millivolts per decade—about as good as it gets. Co–NG comes in at 82 millivolts per decade. The controls lag far behind at 117 and 144. So no, cobalt on graphene doesn't beat platinum in kinetics. But given that it uses orders of magnitude less of a far cheaper metal, the slope is impressive. And unlike plenty of catalysts that look good in acid but fall apart in base, Co–NG holds up in alkaline media too. In one molar sodium hydroxide, it remains more active than either control, so the chemistry of that single-atom site is robust across pH. If you care about intrinsic activity—what each active site is doing—you want to strip away how much material you smeared on the electrode and count the work per atom. Fei and colleagues took a clear stance: treat each cobalt atom as a single catalytic site. With that assumption and an electrochemically measured site density of about 9.7 times ten to the thirteenth sites per square centimeter, they calculated turnover frequencies, or TOFs, across a range of overpotentials. At 50, 100, 150, and 200 millivolts, the catalyst turns over 0.022, 0.101, 0.386, and 1.189 molecules of hydrogen per second per cobalt site. At the thermodynamic potential—zero volts versus the reversible hydrogen electrode—it still moves at 0.0054 per second per site. For context, carefully prepared molybdenum disulfide nanocrystals made under ultra-high vacuum clock in at 0.016 under the same condition, one of the few non-precious benchmarks that beat the cobalt-nitrogen site at that exact point. That comparison is both humbling and encouraging. It says this cobalt atom, sitting in its nitrogen cradle, is in the right ballpark for per-site activity, even if it isn't the absolute champ at zero overpotential. One number that's easy to overlook but crucial for design is how many active sites you can pack in. A perfect platinum surface has on the order of one point five times ten to the fifteenth surface sites per square centimeter. The cobalt-nitrogen graphene here has roughly two orders of magnitude fewer. And yet, because each site is pretty active and because you're using a support where you can process big areas and different formats—including that binder-free paper—the overall currents look healthy. When they benchmarked exchange current density, which is the current at zero overpotential extrapolated from the Tafel line, Co–NG hit about one point two five times ten to the negative four amperes per square centimeter, while nitrogen-doped graphene alone was stuck around eight point three four times ten to the negative seven. That gap drives home the point: the cobalt sites dominate the activity. Stability is where a lot of promising catalysts wash out. Here, the signs are good. After a thousand quick voltammetry cycles in either acid or base, the polarization curves barely budge. In a more practical test—holding the current at 10 milliamps per square centimeter for ten hours—the extra voltage required creeps up by only about 35 millivolts in acid and 17 in base. Post-mortem looks show the cobalt still dispersed as single atoms and the chemical states preserved. The simplest explanation is also the one the spectroscopy suggested from the start: those strong cobalt-nitrogen bonds formed at high temperature lock the atoms in place and keep them from wandering into clusters under operating conditions. Is more cobalt always better? Interestingly, no. When the team varied the cobalt content from as low as about 0.03 percent up to 1.23 percent by atom, the activity rose at first, then leveled off, and finally dropped. That plateau and falloff are consistent with cobalt starting to bump into itself and form clusters once you've saturated the available nitrogen anchoring sites. In other words, the performance lives and dies with the single-atom state. Pack in too much cobalt, lose the isolation, and you're back to average. It's worth pausing on the structural story one more time, because it ties the whole package together. Everything converges: X-ray photoelectron spectroscopy shows the right nitrogen environments and cobalt in an ionic state. Electron microscopy lights up isolated cobalt atoms rather than particles, and extended X-ray absorption fine structure with wavelet analysis fingerprints cobalt-nitrogen coordination in the first shell while ruling out cobalt-cobalt neighbors. Controls made without ammonia, which therefore lack the nitrogen anchors, show cobalt bonded primarily to carbon and perform worse in hydrogen evolution. That structure-function link is the heart of the paper. Not nanoparticles. Not edges alone. Isolated cobalt atoms, stabilized by nitrogen in a graphene lattice, doing the catalytic work. So how does this stack up in the bigger picture? Platinum on carbon still sets the bar, especially in sheer kinetics and in how many active sites you can cram onto a surface. But Fei's cobalt-nitrogen graphene does something that very few non-precious systems manage: it combines low onset overpotential in acid, credible kinetics, activity in base, and real durability, all while using barely more than a rounding error's worth of metal. The practical touches matter too. A synthesis that starts from graphene oxide and a trace cobalt salt, finishes with an ammonia anneal at 750 degrees, and can be shaped into binder-free paper electrodes, doesn't sound like science fiction. It sounds like engineering. If you zoom out, the work points to a broader recipe: build single-atom catalysts by engineering metal-nitrogen coordination on conductive carbons and let the support do more than just hold things together. Nitrogen supplies the seat. The metal sits in it. The graphene conducts the electrons away. You tune the neighborhood—pyridinic here, graphitic there—to set the binding energy of hydrogen where you want it. It's a modular motif, and cobalt is just one guest at the table. Where does it go from here? Carefully, and with experiments. You could test other metals in the same nitrogen seats and ask whether their per-site rates beat cobalt's in acid or base. You could adjust the nitrogen chemistry to tilt the binding energy of hydrogen a little stronger or a little weaker and watch the Tafel slope respond. And you could keep pushing on format—papers, foams, rolled electrodes—to see how these atomically precise sites behave in devices that look like the ones we might actually build. The promise isn't that cobalt-nitrogen graphene is the last word. It's that single atoms, nailed down in the right local chemistry, can make real dents in a reaction we desperately need to scale. For hydrogen, that's a big step. Not a platinum clone, but a smart, frugal alternative that earns its place with atomic efficiency and thoughtful design. As Fei and colleagues showed, sometimes the path to performance isn't more metal. It's a better seat for a single atom to sit in and do its work.