Universal quantum computers offer potentially disruptive solutions to solve currently intractable problems. They provide a tool to artificially reproduce quantum systems with another quantum system that can be programmed at will by making use of phenomena that only occur in the quantum world such as superposition and entanglement. However, modern quantum processors are still relatively simple and their performance is still far from what current classical supercomputers can achieve. Technologies such as ion traps or superconducting qubits are leading the first wave of development with machines of a few tens of quantum bits or qubits1, the basic unit of quantum information processing. These two technologies are attracting the first rounds of public funding at the European level with the Flagship Programme on Quantum Technologies and a significant amount of private funding with superconducting qubit initiatives run by well-established American companies such as IBM, Google or start-ups like IonQ or Rigetti Computing.

However, the current qubit numbers are insufficient to realise quantum computation of significant practical use. To run a simulation of a simple material the predictions set the number of necessary qubits to 10²-10³ units and for an arbitrarily complex one to 10⁶-10⁸ units. Scaling up to a large number of qubits is hence the greatest challenge to fulfill the promises of quantum computing. Current hardware approaches to ion trap and superconducting qubits offer limited prospects for scalability with qubit densities of 1 and 100 qubits/cm², respectively.

Compared to other quantum computing implementations, qubits silicon-based spin qubits offer the best prospects in terms of scalability. The typical footprint of silicon-based spin qubits manufactured with complementary-metal-oxide semiconductor technology is of the order of a few hundred nm².

In addition to their small footprint, Si spin qubits actually show promising performance and a rapid rate of improvement. Recent experimental demonstrations in several academic environments have demonstrated single-qubit gates with fidelities well-above error-correction protocols (>99%17) in cases exceeding those of superconducting qubits, whereas two-qubit gates (F2Q = 98%19) are already above the typical values observed in superconducting systems and are rapidly improving.

In this paper, we will first remind the principle of how to design a qubit based on silicon. Then we will review the current state-of-the-art and will present the architectures to design a large scale quantum computer based on Si spin qubits, their pros and cons regarding variability assumptions and technological achievements will be discussed. We will also show how the recent demonstration of long distance quantum information transfer can be implemented in silicon.