Has a 4 core CPU with a "T" core layout been made?
If not, why not? It seems optimal when attempting to layout in 2D a 4 core CPU cluster with each of the cores directly connected to one another - above the top of the "T" would have a link to connect the end cores & cache fill in above & on the sides making a rectangle to cut. Compare Zen 1 & 2 with the cache in the center splitting the cores: https://i.ytimg.com/vi/NXprMIzv_uw/maxresdefault.jpg The same symmetric splitting with the i7-5960x & Zen 3 & 4 "octo-ring": https://cdn.wccftech.com/wp-content/uploads/2020/11/AMD-Ryzen-5000-Zen-3-Desktop-CPU_Vermeer_Die-Shot_1-scaled.jpg with the ring bringing the need for NUMA. The quad core I7-975 was linear https://www.guru3d.com/miraserver/images/2008/corei7/Nehalem_Die_callout.jpg Quad square configurations appear to have been used for Xeons: https://qph.cf2.quoracdn.net/main-qimg-6762cf8e384ca54f7d7640cadddd511f-pjlq & Athlons: https://static.techspot.com/articles-info/197/images/Image_02-j.webp, but not a "T". If latency being equal between the cores is the reason, then take out middle core & have equidistant tri-cores seesaw that could be linearly chained with the half core excess filled with cache on the ends, but again this does not appear to exist?
The cores don't connect to each other; they connect to the shared cache. Existing layouts are probably near-optimal.
Dumb question. Remember that whatever shape you create has to be one that you can cut out of a wafer. The wafer is a foot across and has many, many tiny copies of your chip etched into the surface. You then use a saw to slice it up into individual devices. Concave shapes are therefore contraindicated.